] :tags
# The metadata that you apply to Amazon Web Services resources to help
# you categorize and organize them. Each tag consists of a key and a
# value, both of which you define. For more information, see [Tagging
# Amazon Web Services Resources][1] in the Amazon Web Services General
# Reference.
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @return [Types::CreateInferenceRecommendationsJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateInferenceRecommendationsJobResponse#job_arn #job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_inference_recommendations_job({
# job_name: "RecommendationJobName", # required
# job_type: "Default", # required, accepts Default, Advanced
# role_arn: "RoleArn", # required
# input_config: { # required
# model_package_version_arn: "ModelPackageArn",
# job_duration_in_seconds: 1,
# traffic_pattern: {
# traffic_type: "PHASES", # accepts PHASES
# phases: [
# {
# initial_number_of_users: 1,
# spawn_rate: 1,
# duration_in_seconds: 1,
# },
# ],
# },
# resource_limit: {
# max_number_of_tests: 1,
# max_parallel_of_tests: 1,
# },
# endpoint_configurations: [
# {
# instance_type: "ml.t2.medium", # required, accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.12xlarge, ml.m5d.24xlarge, ml.c4.large, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.large, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.12xlarge, ml.r5.24xlarge, ml.r5d.large, ml.r5d.xlarge, ml.r5d.2xlarge, ml.r5d.4xlarge, ml.r5d.12xlarge, ml.r5d.24xlarge, ml.inf1.xlarge, ml.inf1.2xlarge, ml.inf1.6xlarge, ml.inf1.24xlarge, ml.c6i.large, ml.c6i.xlarge, ml.c6i.2xlarge, ml.c6i.4xlarge, ml.c6i.8xlarge, ml.c6i.12xlarge, ml.c6i.16xlarge, ml.c6i.24xlarge, ml.c6i.32xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.12xlarge, ml.g5.16xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.p4d.24xlarge, ml.c7g.large, ml.c7g.xlarge, ml.c7g.2xlarge, ml.c7g.4xlarge, ml.c7g.8xlarge, ml.c7g.12xlarge, ml.c7g.16xlarge, ml.m6g.large, ml.m6g.xlarge, ml.m6g.2xlarge, ml.m6g.4xlarge, ml.m6g.8xlarge, ml.m6g.12xlarge, ml.m6g.16xlarge, ml.m6gd.large, ml.m6gd.xlarge, ml.m6gd.2xlarge, ml.m6gd.4xlarge, ml.m6gd.8xlarge, ml.m6gd.12xlarge, ml.m6gd.16xlarge, ml.c6g.large, ml.c6g.xlarge, ml.c6g.2xlarge, ml.c6g.4xlarge, ml.c6g.8xlarge, ml.c6g.12xlarge, ml.c6g.16xlarge, ml.c6gd.large, ml.c6gd.xlarge, ml.c6gd.2xlarge, ml.c6gd.4xlarge, ml.c6gd.8xlarge, ml.c6gd.12xlarge, ml.c6gd.16xlarge, ml.c6gn.large, ml.c6gn.xlarge, ml.c6gn.2xlarge, ml.c6gn.4xlarge, ml.c6gn.8xlarge, ml.c6gn.12xlarge, ml.c6gn.16xlarge, ml.r6g.large, ml.r6g.xlarge, ml.r6g.2xlarge, ml.r6g.4xlarge, ml.r6g.8xlarge, ml.r6g.12xlarge, ml.r6g.16xlarge, ml.r6gd.large, ml.r6gd.xlarge, ml.r6gd.2xlarge, ml.r6gd.4xlarge, ml.r6gd.8xlarge, ml.r6gd.12xlarge, ml.r6gd.16xlarge, ml.p4de.24xlarge, ml.trn1.2xlarge, ml.trn1.32xlarge, ml.inf2.xlarge, ml.inf2.8xlarge, ml.inf2.24xlarge, ml.inf2.48xlarge
# inference_specification_name: "InferenceSpecificationName",
# environment_parameter_ranges: {
# categorical_parameter_ranges: [
# {
# name: "String64", # required
# value: ["String128"], # required
# },
# ],
# },
# },
# ],
# volume_kms_key_id: "KmsKeyId",
# container_config: {
# domain: "String",
# task: "String",
# framework: "String",
# framework_version: "String",
# payload_config: {
# sample_payload_url: "String",
# supported_content_types: ["String"],
# },
# nearest_model_name: "String",
# supported_instance_types: ["String"],
# data_input_config: "RecommendationJobDataInputConfig",
# },
# endpoints: [
# {
# endpoint_name: "EndpointName", # required
# },
# ],
# vpc_config: {
# security_group_ids: ["RecommendationJobVpcSecurityGroupId"], # required
# subnets: ["RecommendationJobVpcSubnetId"], # required
# },
# model_name: "ModelName",
# },
# job_description: "RecommendationJobDescription",
# stopping_conditions: {
# max_invocations: 1,
# model_latency_thresholds: [
# {
# percentile: "String64",
# value_in_milliseconds: 1,
# },
# ],
# },
# output_config: {
# kms_key_id: "KmsKeyId",
# compiled_output_config: {
# s3_output_uri: "S3Uri",
# },
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateInferenceRecommendationsJob AWS API Documentation
#
# @overload create_inference_recommendations_job(params = {})
# @param [Hash] params ({})
def create_inference_recommendations_job(params = {}, options = {})
req = build_request(:create_inference_recommendations_job, params)
req.send_request(options)
end
# Creates a job that uses workers to label the data objects in your
# input dataset. You can use the labeled data to train machine learning
# models.
#
# You can select your workforce from one of three providers:
#
# * A private workforce that you create. It can include employees,
# contractors, and outside experts. Use a private workforce when want
# the data to stay within your organization or when a specific set of
# skills is required.
#
# * One or more vendors that you select from the Amazon Web Services
# Marketplace. Vendors provide expertise in specific areas.
#
# * The Amazon Mechanical Turk workforce. This is the largest workforce,
# but it should only be used for public data or data that has been
# stripped of any personally identifiable information.
#
# You can also use *automated data labeling* to reduce the number of
# data objects that need to be labeled by a human. Automated data
# labeling uses *active learning* to determine if a data object can be
# labeled by machine or if it needs to be sent to a human worker. For
# more information, see [Using Automated Data Labeling][1].
#
# The data objects to be labeled are contained in an Amazon S3 bucket.
# You create a *manifest file* that describes the location of each
# object. For more information, see [Using Input and Output Data][2].
#
# The output can be used as the manifest file for another labeling job
# or as training data for your machine learning models.
#
# You can use this operation to create a static labeling job or a
# streaming labeling job. A static labeling job stops if all data
# objects in the input manifest file identified in `ManifestS3Uri` have
# been labeled. A streaming labeling job runs perpetually until it is
# manually stopped, or remains idle for 10 days. You can send new data
# objects to an active (`InProgress`) streaming labeling job in real
# time. To learn how to create a static labeling job, see [Create a
# Labeling Job (API) ][3] in the Amazon SageMaker Developer Guide. To
# learn how to create a streaming labeling job, see [Create a Streaming
# Labeling Job][4].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-automated-labeling.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-data.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-create-labeling-job-api.html
# [4]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-streaming-create-job.html
#
# @option params [required, String] :labeling_job_name
# The name of the labeling job. This name is used to identify the job in
# a list of labeling jobs. Labeling job names must be unique within an
# Amazon Web Services account and region. `LabelingJobName` is not case
# sensitive. For example, Example-job and example-job are considered the
# same labeling job name by Ground Truth.
#
# @option params [required, String] :label_attribute_name
# The attribute name to use for the label in the output manifest file.
# This is the key for the key/value pair formed with the label that a
# worker assigns to the object. The `LabelAttributeName` must meet the
# following requirements.
#
# * The name can't end with "-metadata".
#
# * If you are using one of the following [built-in task types][1], the
# attribute name *must* end with "-ref". If the task type you are
# using is not listed below, the attribute name *must not* end with
# "-ref".
#
# * Image semantic segmentation (`SemanticSegmentation)`, and
# adjustment (`AdjustmentSemanticSegmentation`) and verification
# (`VerificationSemanticSegmentation`) labeling jobs for this task
# type.
#
# * Video frame object detection (`VideoObjectDetection`), and
# adjustment and verification (`AdjustmentVideoObjectDetection`)
# labeling jobs for this task type.
#
# * Video frame object tracking (`VideoObjectTracking`), and
# adjustment and verification (`AdjustmentVideoObjectTracking`)
# labeling jobs for this task type.
#
# * 3D point cloud semantic segmentation
# (`3DPointCloudSemanticSegmentation`), and adjustment and
# verification (`Adjustment3DPointCloudSemanticSegmentation`)
# labeling jobs for this task type.
#
# * 3D point cloud object tracking (`3DPointCloudObjectTracking`), and
# adjustment and verification
# (`Adjustment3DPointCloudObjectTracking`) labeling jobs for this
# task type.
#
#
#
# If you are creating an adjustment or verification labeling job, you
# must use a *different* `LabelAttributeName` than the one used in the
# original labeling job. The original labeling job is the Ground Truth
# labeling job that produced the labels that you want verified or
# adjusted. To learn more about adjustment and verification labeling
# jobs, see [Verify and Adjust Labels][2].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-task-types.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-verification-data.html
#
# @option params [required, Types::LabelingJobInputConfig] :input_config
# Input data for the labeling job, such as the Amazon S3 location of the
# data objects and the location of the manifest file that describes the
# data objects.
#
# You must specify at least one of the following: `S3DataSource` or
# `SnsDataSource`.
#
# * Use `SnsDataSource` to specify an SNS input topic for a streaming
# labeling job. If you do not specify and SNS input topic ARN, Ground
# Truth will create a one-time labeling job that stops after all data
# objects in the input manifest file have been labeled.
#
# * Use `S3DataSource` to specify an input manifest file for both
# streaming and one-time labeling jobs. Adding an `S3DataSource` is
# optional if you use `SnsDataSource` to create a streaming labeling
# job.
#
# If you use the Amazon Mechanical Turk workforce, your input data
# should not include confidential information, personal information or
# protected health information. Use `ContentClassifiers` to specify that
# your data is free of personally identifiable information and adult
# content.
#
# @option params [required, Types::LabelingJobOutputConfig] :output_config
# The location of the output data and the Amazon Web Services Key
# Management Service key ID for the key used to encrypt the output data,
# if any.
#
# @option params [required, String] :role_arn
# The Amazon Resource Number (ARN) that Amazon SageMaker assumes to
# perform tasks on your behalf during data labeling. You must grant this
# role the necessary permissions so that Amazon SageMaker can
# successfully complete data labeling.
#
# @option params [String] :label_category_config_s3_uri
# The S3 URI of the file, referred to as a *label category configuration
# file*, that defines the categories used to label the data objects.
#
# For 3D point cloud and video frame task types, you can add label
# category attributes and frame attributes to your label category
# configuration file. To learn how, see [Create a Labeling Category
# Configuration File for 3D Point Cloud Labeling Jobs][1].
#
# For named entity recognition jobs, in addition to `"labels"`, you must
# provide worker instructions in the label category configuration file
# using the `"instructions"` parameter: `"instructions":
# \{"shortInstruction":"Add header
Add Instructions
",
# "fullInstruction":"Add additional instructions.
"\}`. For
# details and an example, see [Create a Named Entity Recognition
# Labeling Job (API) ][2].
#
# For all other [built-in task types][3] and [custom tasks][4], your
# label category configuration file must be a JSON file in the following
# format. Identify the labels you want to use by replacing `label_1`,
# `label_2`,`...`,`label_n` with your label categories.
#
# `\{ `
#
# `"document-version": "2018-11-28",`
#
# `"labels": [\{"label": "label_1"\},\{"label":
# "label_2"\},...\{"label": "label_n"\}]`
#
# `\}`
#
# Note the following about the label category configuration file:
#
# * For image classification and text classification (single and
# multi-label) you must specify at least two label categories. For all
# other task types, the minimum number of label categories required is
# one.
#
# * Each label category must be unique, you cannot specify duplicate
# label categories.
#
# * If you create a 3D point cloud or video frame adjustment or
# verification labeling job, you must include
# `auditLabelAttributeName` in the label category configuration. Use
# this parameter to enter the [ `LabelAttributeName` ][5] of the
# labeling job you want to adjust or verify annotations of.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-point-cloud-label-category-config.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-named-entity-recg.html#sms-creating-ner-api
# [3]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-task-types.html
# [4]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-custom-templates.html
# [5]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateLabelingJob.html#sagemaker-CreateLabelingJob-request-LabelAttributeName
#
# @option params [Types::LabelingJobStoppingConditions] :stopping_conditions
# A set of conditions for stopping the labeling job. If any of the
# conditions are met, the job is automatically stopped. You can use
# these conditions to control the cost of data labeling.
#
# @option params [Types::LabelingJobAlgorithmsConfig] :labeling_job_algorithms_config
# Configures the information required to perform automated data
# labeling.
#
# @option params [required, Types::HumanTaskConfig] :human_task_config
# Configures the labeling task and how it is presented to workers;
# including, but not limited to price, keywords, and batch size (task
# count).
#
# @option params [Array] :tags
# An array of key/value pairs. For more information, see [Using Cost
# Allocation Tags][1] in the *Amazon Web Services Billing and Cost
# Management User Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-what
#
# @return [Types::CreateLabelingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateLabelingJobResponse#labeling_job_arn #labeling_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_labeling_job({
# labeling_job_name: "LabelingJobName", # required
# label_attribute_name: "LabelAttributeName", # required
# input_config: { # required
# data_source: { # required
# s3_data_source: {
# manifest_s3_uri: "S3Uri", # required
# },
# sns_data_source: {
# sns_topic_arn: "SnsTopicArn", # required
# },
# },
# data_attributes: {
# content_classifiers: ["FreeOfPersonallyIdentifiableInformation"], # accepts FreeOfPersonallyIdentifiableInformation, FreeOfAdultContent
# },
# },
# output_config: { # required
# s3_output_path: "S3Uri", # required
# kms_key_id: "KmsKeyId",
# sns_topic_arn: "SnsTopicArn",
# },
# role_arn: "RoleArn", # required
# label_category_config_s3_uri: "S3Uri",
# stopping_conditions: {
# max_human_labeled_object_count: 1,
# max_percentage_of_input_dataset_labeled: 1,
# },
# labeling_job_algorithms_config: {
# labeling_job_algorithm_specification_arn: "LabelingJobAlgorithmSpecificationArn", # required
# initial_active_learning_model_arn: "ModelArn",
# labeling_job_resource_config: {
# volume_kms_key_id: "KmsKeyId",
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# },
# human_task_config: { # required
# workteam_arn: "WorkteamArn", # required
# ui_config: { # required
# ui_template_s3_uri: "S3Uri",
# human_task_ui_arn: "HumanTaskUiArn",
# },
# pre_human_task_lambda_arn: "LambdaFunctionArn", # required
# task_keywords: ["TaskKeyword"],
# task_title: "TaskTitle", # required
# task_description: "TaskDescription", # required
# number_of_human_workers_per_data_object: 1, # required
# task_time_limit_in_seconds: 1, # required
# task_availability_lifetime_in_seconds: 1,
# max_concurrent_task_count: 1,
# annotation_consolidation_config: { # required
# annotation_consolidation_lambda_arn: "LambdaFunctionArn", # required
# },
# public_workforce_task_price: {
# amount_in_usd: {
# dollars: 1,
# cents: 1,
# tenth_fractions_of_a_cent: 1,
# },
# },
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.labeling_job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateLabelingJob AWS API Documentation
#
# @overload create_labeling_job(params = {})
# @param [Hash] params ({})
def create_labeling_job(params = {}, options = {})
req = build_request(:create_labeling_job, params)
req.send_request(options)
end
# Creates a model in SageMaker. In the request, you name the model and
# describe a primary container. For the primary container, you specify
# the Docker image that contains inference code, artifacts (from prior
# training), and a custom environment map that the inference code uses
# when you deploy the model for predictions.
#
# Use this API to create a model if you want to use SageMaker hosting
# services or run a batch transform job.
#
# To host your model, you create an endpoint configuration with the
# `CreateEndpointConfig` API, and then create an endpoint with the
# `CreateEndpoint` API. SageMaker then deploys all of the containers
# that you defined for the model in the hosting environment.
#
# For an example that calls this method when deploying a model to
# SageMaker hosting services, see [Create a Model (Amazon Web Services
# SDK for Python (Boto 3)).][1]
#
# To run a batch transform using your model, you start a job with the
# `CreateTransformJob` API. SageMaker uses your model and your dataset
# to get inferences which are then saved to a specified S3 location.
#
# In the request, you also provide an IAM role that SageMaker can assume
# to access model artifacts and docker image for deployment on ML
# compute hosting instances or for batch transform jobs. In addition,
# you also use the IAM role to manage permissions the inference code
# needs. For example, if the inference code access any other Amazon Web
# Services resources, you grant necessary permissions via this role.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/realtime-endpoints-deployment.html#realtime-endpoints-deployment-create-model
#
# @option params [required, String] :model_name
# The name of the new model.
#
# @option params [Types::ContainerDefinition] :primary_container
# The location of the primary docker image containing inference code,
# associated artifacts, and custom environment map that the inference
# code uses when the model is deployed for predictions.
#
# @option params [Array] :containers
# Specifies the containers in the inference pipeline.
#
# @option params [Types::InferenceExecutionConfig] :inference_execution_config
# Specifies details of how containers in a multi-container endpoint are
# called.
#
# @option params [required, String] :execution_role_arn
# The Amazon Resource Name (ARN) of the IAM role that SageMaker can
# assume to access model artifacts and docker image for deployment on ML
# compute instances or for batch transform jobs. Deploying on ML compute
# instances is part of model hosting. For more information, see
# [SageMaker Roles][1].
#
# To be able to pass this role to SageMaker, the caller of this API must
# have the `iam:PassRole` permission.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html
#
# @option params [Array] :tags
# An array of key-value pairs. You can use tags to categorize your
# Amazon Web Services resources in different ways, for example, by
# purpose, owner, or environment. For more information, see [Tagging
# Amazon Web Services Resources][1].
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @option params [Types::VpcConfig] :vpc_config
# A [VpcConfig][1] object that specifies the VPC that you want your
# model to connect to. Control access to and from your model container
# by configuring the VPC. `VpcConfig` is used in hosting services and in
# batch transform. For more information, see [Protect Endpoints by Using
# an Amazon Virtual Private Cloud][2] and [Protect Data in Batch
# Transform Jobs by Using an Amazon Virtual Private Cloud][3].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_VpcConfig.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/host-vpc.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/dg/batch-vpc.html
#
# @option params [Boolean] :enable_network_isolation
# Isolates the model container. No inbound or outbound network calls can
# be made to or from the model container.
#
# @return [Types::CreateModelOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelOutput#model_arn #model_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model({
# model_name: "ModelName", # required
# primary_container: {
# container_hostname: "ContainerHostname",
# image: "ContainerImage",
# image_config: {
# repository_access_mode: "Platform", # required, accepts Platform, Vpc
# repository_auth_config: {
# repository_credentials_provider_arn: "RepositoryCredentialsProviderArn", # required
# },
# },
# mode: "SingleModel", # accepts SingleModel, MultiModel
# model_data_url: "Url",
# environment: {
# "EnvironmentKey" => "EnvironmentValue",
# },
# model_package_name: "VersionedArnOrName",
# inference_specification_name: "InferenceSpecificationName",
# multi_model_config: {
# model_cache_setting: "Enabled", # accepts Enabled, Disabled
# },
# },
# containers: [
# {
# container_hostname: "ContainerHostname",
# image: "ContainerImage",
# image_config: {
# repository_access_mode: "Platform", # required, accepts Platform, Vpc
# repository_auth_config: {
# repository_credentials_provider_arn: "RepositoryCredentialsProviderArn", # required
# },
# },
# mode: "SingleModel", # accepts SingleModel, MultiModel
# model_data_url: "Url",
# environment: {
# "EnvironmentKey" => "EnvironmentValue",
# },
# model_package_name: "VersionedArnOrName",
# inference_specification_name: "InferenceSpecificationName",
# multi_model_config: {
# model_cache_setting: "Enabled", # accepts Enabled, Disabled
# },
# },
# ],
# inference_execution_config: {
# mode: "Serial", # required, accepts Serial, Direct
# },
# execution_role_arn: "RoleArn", # required
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# enable_network_isolation: false,
# })
#
# @example Response structure
#
# resp.model_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModel AWS API Documentation
#
# @overload create_model(params = {})
# @param [Hash] params ({})
def create_model(params = {}, options = {})
req = build_request(:create_model, params)
req.send_request(options)
end
# Creates the definition for a model bias job.
#
# @option params [required, String] :job_definition_name
# The name of the bias job definition. The name must be unique within an
# Amazon Web Services Region in the Amazon Web Services account.
#
# @option params [Types::ModelBiasBaselineConfig] :model_bias_baseline_config
# The baseline configuration for a model bias job.
#
# @option params [required, Types::ModelBiasAppSpecification] :model_bias_app_specification
# Configures the model bias job to run a specified Docker container
# image.
#
# @option params [required, Types::ModelBiasJobInput] :model_bias_job_input
# Inputs for the model bias job.
#
# @option params [required, Types::MonitoringOutputConfig] :model_bias_job_output_config
# The output configuration for monitoring jobs.
#
# @option params [required, Types::MonitoringResources] :job_resources
# Identifies the resources to deploy for a monitoring job.
#
# @option params [Types::MonitoringNetworkConfig] :network_config
# Networking options for a model bias job.
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) of an IAM role that Amazon SageMaker
# can assume to perform tasks on your behalf.
#
# @option params [Types::MonitoringStoppingCondition] :stopping_condition
# A time limit for how long the monitoring job is allowed to run before
# stopping.
#
# @option params [Array] :tags
# (Optional) An array of key-value pairs. For more information, see
# [Using Cost Allocation Tags][1] in the *Amazon Web Services Billing
# and Cost Management User Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-whatURL
#
# @return [Types::CreateModelBiasJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelBiasJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_bias_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# model_bias_baseline_config: {
# baselining_job_name: "ProcessingJobName",
# constraints_resource: {
# s3_uri: "S3Uri",
# },
# },
# model_bias_app_specification: { # required
# image_uri: "ImageUri", # required
# config_uri: "S3Uri", # required
# environment: {
# "ProcessingEnvironmentKey" => "ProcessingEnvironmentValue",
# },
# },
# model_bias_job_input: { # required
# endpoint_input: {
# endpoint_name: "EndpointName", # required
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# batch_transform_input: {
# data_captured_destination_s3_uri: "DestinationS3Uri", # required
# dataset_format: { # required
# csv: {
# header: false,
# },
# json: {
# line: false,
# },
# parquet: {
# },
# },
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# ground_truth_s3_input: { # required
# s3_uri: "MonitoringS3Uri",
# },
# },
# model_bias_job_output_config: { # required
# monitoring_outputs: [ # required
# {
# s3_output: { # required
# s3_uri: "MonitoringS3Uri", # required
# local_path: "ProcessingLocalPath", # required
# s3_upload_mode: "Continuous", # accepts Continuous, EndOfJob
# },
# },
# ],
# kms_key_id: "KmsKeyId",
# },
# job_resources: { # required
# cluster_config: { # required
# instance_count: 1, # required
# instance_type: "ml.t3.medium", # required, accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# network_config: {
# enable_inter_container_traffic_encryption: false,
# enable_network_isolation: false,
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# role_arn: "RoleArn", # required
# stopping_condition: {
# max_runtime_in_seconds: 1, # required
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelBiasJobDefinition AWS API Documentation
#
# @overload create_model_bias_job_definition(params = {})
# @param [Hash] params ({})
def create_model_bias_job_definition(params = {}, options = {})
req = build_request(:create_model_bias_job_definition, params)
req.send_request(options)
end
# Creates an Amazon SageMaker Model Card.
#
# For information about how to use model cards, see [Amazon SageMaker
# Model Card][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/model-cards.html
#
# @option params [required, String] :model_card_name
# The unique name of the model card.
#
# @option params [Types::ModelCardSecurityConfig] :security_config
# An optional Key Management Service key to encrypt, decrypt, and
# re-encrypt model card content for regulated workloads with highly
# sensitive data.
#
# @option params [required, String] :content
# The content of the model card. Content must be in [model card JSON
# schema][1] and provided as a string.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/model-cards.html#model-cards-json-schema
#
# @option params [required, String] :model_card_status
# The approval status of the model card within your organization.
# Different organizations might have different criteria for model card
# review and approval.
#
# * `Draft`: The model card is a work in progress.
#
# * `PendingReview`: The model card is pending review.
#
# * `Approved`: The model card is approved.
#
# * `Archived`: The model card is archived. No more updates should be
# made to the model card, but it can still be exported.
#
# @option params [Array] :tags
# Key-value pairs used to manage metadata for model cards.
#
# @return [Types::CreateModelCardResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelCardResponse#model_card_arn #model_card_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_card({
# model_card_name: "EntityName", # required
# security_config: {
# kms_key_id: "KmsKeyId",
# },
# content: "ModelCardContent", # required
# model_card_status: "Draft", # required, accepts Draft, PendingReview, Approved, Archived
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.model_card_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelCard AWS API Documentation
#
# @overload create_model_card(params = {})
# @param [Hash] params ({})
def create_model_card(params = {}, options = {})
req = build_request(:create_model_card, params)
req.send_request(options)
end
# Creates an Amazon SageMaker Model Card export job.
#
# @option params [required, String] :model_card_name
# The name of the model card to export.
#
# @option params [Integer] :model_card_version
# The version of the model card to export. If a version is not provided,
# then the latest version of the model card is exported.
#
# @option params [required, String] :model_card_export_job_name
# The name of the model card export job.
#
# @option params [required, Types::ModelCardExportOutputConfig] :output_config
# The model card output configuration that specifies the Amazon S3 path
# for exporting.
#
# @return [Types::CreateModelCardExportJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelCardExportJobResponse#model_card_export_job_arn #model_card_export_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_card_export_job({
# model_card_name: "EntityName", # required
# model_card_version: 1,
# model_card_export_job_name: "EntityName", # required
# output_config: { # required
# s3_output_path: "S3Uri", # required
# },
# })
#
# @example Response structure
#
# resp.model_card_export_job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelCardExportJob AWS API Documentation
#
# @overload create_model_card_export_job(params = {})
# @param [Hash] params ({})
def create_model_card_export_job(params = {}, options = {})
req = build_request(:create_model_card_export_job, params)
req.send_request(options)
end
# Creates the definition for a model explainability job.
#
# @option params [required, String] :job_definition_name
# The name of the model explainability job definition. The name must be
# unique within an Amazon Web Services Region in the Amazon Web Services
# account.
#
# @option params [Types::ModelExplainabilityBaselineConfig] :model_explainability_baseline_config
# The baseline configuration for a model explainability job.
#
# @option params [required, Types::ModelExplainabilityAppSpecification] :model_explainability_app_specification
# Configures the model explainability job to run a specified Docker
# container image.
#
# @option params [required, Types::ModelExplainabilityJobInput] :model_explainability_job_input
# Inputs for the model explainability job.
#
# @option params [required, Types::MonitoringOutputConfig] :model_explainability_job_output_config
# The output configuration for monitoring jobs.
#
# @option params [required, Types::MonitoringResources] :job_resources
# Identifies the resources to deploy for a monitoring job.
#
# @option params [Types::MonitoringNetworkConfig] :network_config
# Networking options for a model explainability job.
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) of an IAM role that Amazon SageMaker
# can assume to perform tasks on your behalf.
#
# @option params [Types::MonitoringStoppingCondition] :stopping_condition
# A time limit for how long the monitoring job is allowed to run before
# stopping.
#
# @option params [Array] :tags
# (Optional) An array of key-value pairs. For more information, see
# [Using Cost Allocation Tags][1] in the *Amazon Web Services Billing
# and Cost Management User Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-whatURL
#
# @return [Types::CreateModelExplainabilityJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelExplainabilityJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_explainability_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# model_explainability_baseline_config: {
# baselining_job_name: "ProcessingJobName",
# constraints_resource: {
# s3_uri: "S3Uri",
# },
# },
# model_explainability_app_specification: { # required
# image_uri: "ImageUri", # required
# config_uri: "S3Uri", # required
# environment: {
# "ProcessingEnvironmentKey" => "ProcessingEnvironmentValue",
# },
# },
# model_explainability_job_input: { # required
# endpoint_input: {
# endpoint_name: "EndpointName", # required
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# batch_transform_input: {
# data_captured_destination_s3_uri: "DestinationS3Uri", # required
# dataset_format: { # required
# csv: {
# header: false,
# },
# json: {
# line: false,
# },
# parquet: {
# },
# },
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# },
# model_explainability_job_output_config: { # required
# monitoring_outputs: [ # required
# {
# s3_output: { # required
# s3_uri: "MonitoringS3Uri", # required
# local_path: "ProcessingLocalPath", # required
# s3_upload_mode: "Continuous", # accepts Continuous, EndOfJob
# },
# },
# ],
# kms_key_id: "KmsKeyId",
# },
# job_resources: { # required
# cluster_config: { # required
# instance_count: 1, # required
# instance_type: "ml.t3.medium", # required, accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# network_config: {
# enable_inter_container_traffic_encryption: false,
# enable_network_isolation: false,
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# role_arn: "RoleArn", # required
# stopping_condition: {
# max_runtime_in_seconds: 1, # required
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelExplainabilityJobDefinition AWS API Documentation
#
# @overload create_model_explainability_job_definition(params = {})
# @param [Hash] params ({})
def create_model_explainability_job_definition(params = {}, options = {})
req = build_request(:create_model_explainability_job_definition, params)
req.send_request(options)
end
# Creates a model package that you can use to create SageMaker models or
# list on Amazon Web Services Marketplace, or a versioned model that is
# part of a model group. Buyers can subscribe to model packages listed
# on Amazon Web Services Marketplace to create models in SageMaker.
#
# To create a model package by specifying a Docker container that
# contains your inference code and the Amazon S3 location of your model
# artifacts, provide values for `InferenceSpecification`. To create a
# model from an algorithm resource that you created or subscribed to in
# Amazon Web Services Marketplace, provide a value for
# `SourceAlgorithmSpecification`.
#
# There are two types of model packages:
#
# * Versioned - a model that is part of a model group in the model
# registry.
#
# * Unversioned - a model package that is not part of a model group.
#
#
#
# @option params [String] :model_package_name
# The name of the model package. The name must have 1 to 63 characters.
# Valid characters are a-z, A-Z, 0-9, and - (hyphen).
#
# This parameter is required for unversioned models. It is not
# applicable to versioned models.
#
# @option params [String] :model_package_group_name
# The name or Amazon Resource Name (ARN) of the model package group that
# this model version belongs to.
#
# This parameter is required for versioned models, and does not apply to
# unversioned models.
#
# @option params [String] :model_package_description
# A description of the model package.
#
# @option params [Types::InferenceSpecification] :inference_specification
# Specifies details about inference jobs that can be run with models
# based on this model package, including the following:
#
# * The Amazon ECR paths of containers that contain the inference code
# and model artifacts.
#
# * The instance types that the model package supports for transform
# jobs and real-time endpoints used for inference.
#
# * The input and output content formats that the model package supports
# for inference.
#
# @option params [Types::ModelPackageValidationSpecification] :validation_specification
# Specifies configurations for one or more transform jobs that SageMaker
# runs to test the model package.
#
# @option params [Types::SourceAlgorithmSpecification] :source_algorithm_specification
# Details about the algorithm that was used to create the model package.
#
# @option params [Boolean] :certify_for_marketplace
# Whether to certify the model package for listing on Amazon Web
# Services Marketplace.
#
# This parameter is optional for unversioned models, and does not apply
# to versioned models.
#
# @option params [Array] :tags
# A list of key value pairs associated with the model. For more
# information, see [Tagging Amazon Web Services resources][1] in the
# *Amazon Web Services General Reference Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @option params [String] :model_approval_status
# Whether the model is approved for deployment.
#
# This parameter is optional for versioned models, and does not apply to
# unversioned models.
#
# For versioned models, the value of this parameter must be set to
# `Approved` to deploy the model.
#
# @option params [Types::MetadataProperties] :metadata_properties
# Metadata properties of the tracking entity, trial, or trial component.
#
# @option params [Types::ModelMetrics] :model_metrics
# A structure that contains model metrics reports.
#
# @option params [String] :client_token
# A unique token that guarantees that the call to this API is
# idempotent.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @option params [Hash] :customer_metadata_properties
# The metadata properties associated with the model package versions.
#
# @option params [Types::DriftCheckBaselines] :drift_check_baselines
# Represents the drift check baselines that can be used when the model
# monitor is set using the model package. For more information, see the
# topic on [Drift Detection against Previous Baselines in SageMaker
# Pipelines][1] in the *Amazon SageMaker Developer Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/pipelines-quality-clarify-baseline-lifecycle.html#pipelines-quality-clarify-baseline-drift-detection
#
# @option params [String] :domain
# The machine learning domain of your model package and its components.
# Common machine learning domains include computer vision and natural
# language processing.
#
# @option params [String] :task
# The machine learning task your model package accomplishes. Common
# machine learning tasks include object detection and image
# classification. The following tasks are supported by Inference
# Recommender: `"IMAGE_CLASSIFICATION"` \| `"OBJECT_DETECTION"` \|
# `"TEXT_GENERATION"` \|`"IMAGE_SEGMENTATION"` \| `"FILL_MASK"` \|
# `"CLASSIFICATION"` \| `"REGRESSION"` \| `"OTHER"`.
#
# Specify "OTHER" if none of the tasks listed fit your use case.
#
# @option params [String] :sample_payload_url
# The Amazon Simple Storage Service (Amazon S3) path where the sample
# payload is stored. This path must point to a single gzip compressed
# tar archive (.tar.gz suffix). This archive can hold multiple files
# that are all equally used in the load test. Each file in the archive
# must satisfy the size constraints of the [InvokeEndpoint][1] call.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_runtime_InvokeEndpoint.html#API_runtime_InvokeEndpoint_RequestSyntax
#
# @option params [Array] :additional_inference_specifications
# An array of additional Inference Specification objects. Each
# additional Inference Specification specifies artifacts based on this
# model package that can be used on inference endpoints. Generally used
# with SageMaker Neo to store the compiled artifacts.
#
# @return [Types::CreateModelPackageOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelPackageOutput#model_package_arn #model_package_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_package({
# model_package_name: "EntityName",
# model_package_group_name: "ArnOrName",
# model_package_description: "EntityDescription",
# inference_specification: {
# containers: [ # required
# {
# container_hostname: "ContainerHostname",
# image: "ContainerImage", # required
# image_digest: "ImageDigest",
# model_data_url: "Url",
# product_id: "ProductId",
# environment: {
# "EnvironmentKey" => "EnvironmentValue",
# },
# model_input: {
# data_input_config: "DataInputConfig", # required
# },
# framework: "String",
# framework_version: "ModelPackageFrameworkVersion",
# nearest_model_name: "String",
# },
# ],
# supported_transform_instance_types: ["ml.m4.xlarge"], # accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# supported_realtime_inference_instance_types: ["ml.t2.medium"], # accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.12xlarge, ml.m5d.24xlarge, ml.c4.large, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.large, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.12xlarge, ml.r5.24xlarge, ml.r5d.large, ml.r5d.xlarge, ml.r5d.2xlarge, ml.r5d.4xlarge, ml.r5d.12xlarge, ml.r5d.24xlarge, ml.inf1.xlarge, ml.inf1.2xlarge, ml.inf1.6xlarge, ml.inf1.24xlarge, ml.c6i.large, ml.c6i.xlarge, ml.c6i.2xlarge, ml.c6i.4xlarge, ml.c6i.8xlarge, ml.c6i.12xlarge, ml.c6i.16xlarge, ml.c6i.24xlarge, ml.c6i.32xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.12xlarge, ml.g5.16xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.p4d.24xlarge, ml.c7g.large, ml.c7g.xlarge, ml.c7g.2xlarge, ml.c7g.4xlarge, ml.c7g.8xlarge, ml.c7g.12xlarge, ml.c7g.16xlarge, ml.m6g.large, ml.m6g.xlarge, ml.m6g.2xlarge, ml.m6g.4xlarge, ml.m6g.8xlarge, ml.m6g.12xlarge, ml.m6g.16xlarge, ml.m6gd.large, ml.m6gd.xlarge, ml.m6gd.2xlarge, ml.m6gd.4xlarge, ml.m6gd.8xlarge, ml.m6gd.12xlarge, ml.m6gd.16xlarge, ml.c6g.large, ml.c6g.xlarge, ml.c6g.2xlarge, ml.c6g.4xlarge, ml.c6g.8xlarge, ml.c6g.12xlarge, ml.c6g.16xlarge, ml.c6gd.large, ml.c6gd.xlarge, ml.c6gd.2xlarge, ml.c6gd.4xlarge, ml.c6gd.8xlarge, ml.c6gd.12xlarge, ml.c6gd.16xlarge, ml.c6gn.large, ml.c6gn.xlarge, ml.c6gn.2xlarge, ml.c6gn.4xlarge, ml.c6gn.8xlarge, ml.c6gn.12xlarge, ml.c6gn.16xlarge, ml.r6g.large, ml.r6g.xlarge, ml.r6g.2xlarge, ml.r6g.4xlarge, ml.r6g.8xlarge, ml.r6g.12xlarge, ml.r6g.16xlarge, ml.r6gd.large, ml.r6gd.xlarge, ml.r6gd.2xlarge, ml.r6gd.4xlarge, ml.r6gd.8xlarge, ml.r6gd.12xlarge, ml.r6gd.16xlarge, ml.p4de.24xlarge, ml.trn1.2xlarge, ml.trn1.32xlarge, ml.inf2.xlarge, ml.inf2.8xlarge, ml.inf2.24xlarge, ml.inf2.48xlarge
# supported_content_types: ["ContentType"], # required
# supported_response_mime_types: ["ResponseMIMEType"], # required
# },
# validation_specification: {
# validation_role: "RoleArn", # required
# validation_profiles: [ # required
# {
# profile_name: "EntityName", # required
# transform_job_definition: { # required
# max_concurrent_transforms: 1,
# max_payload_in_mb: 1,
# batch_strategy: "MultiRecord", # accepts MultiRecord, SingleRecord
# environment: {
# "TransformEnvironmentKey" => "TransformEnvironmentValue",
# },
# transform_input: { # required
# data_source: { # required
# s3_data_source: { # required
# s3_data_type: "ManifestFile", # required, accepts ManifestFile, S3Prefix, AugmentedManifestFile
# s3_uri: "S3Uri", # required
# },
# },
# content_type: "ContentType",
# compression_type: "None", # accepts None, Gzip
# split_type: "None", # accepts None, Line, RecordIO, TFRecord
# },
# transform_output: { # required
# s3_output_path: "S3Uri", # required
# accept: "Accept",
# assemble_with: "None", # accepts None, Line
# kms_key_id: "KmsKeyId",
# },
# transform_resources: { # required
# instance_type: "ml.m4.xlarge", # required, accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# instance_count: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# },
# ],
# },
# source_algorithm_specification: {
# source_algorithms: [ # required
# {
# model_data_url: "Url",
# algorithm_name: "ArnOrName", # required
# },
# ],
# },
# certify_for_marketplace: false,
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# model_approval_status: "Approved", # accepts Approved, Rejected, PendingManualApproval
# metadata_properties: {
# commit_id: "MetadataPropertyValue",
# repository: "MetadataPropertyValue",
# generated_by: "MetadataPropertyValue",
# project_id: "MetadataPropertyValue",
# },
# model_metrics: {
# model_quality: {
# statistics: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# model_data_quality: {
# statistics: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# bias: {
# report: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# pre_training_report: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# post_training_report: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# explainability: {
# report: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# },
# client_token: "ClientToken",
# customer_metadata_properties: {
# "CustomerMetadataKey" => "CustomerMetadataValue",
# },
# drift_check_baselines: {
# bias: {
# config_file: {
# content_type: "ContentType",
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# pre_training_constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# post_training_constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# explainability: {
# constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# config_file: {
# content_type: "ContentType",
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# model_quality: {
# statistics: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# model_data_quality: {
# statistics: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# constraints: {
# content_type: "ContentType", # required
# content_digest: "ContentDigest",
# s3_uri: "S3Uri", # required
# },
# },
# },
# domain: "String",
# task: "String",
# sample_payload_url: "S3Uri",
# additional_inference_specifications: [
# {
# name: "EntityName", # required
# description: "EntityDescription",
# containers: [ # required
# {
# container_hostname: "ContainerHostname",
# image: "ContainerImage", # required
# image_digest: "ImageDigest",
# model_data_url: "Url",
# product_id: "ProductId",
# environment: {
# "EnvironmentKey" => "EnvironmentValue",
# },
# model_input: {
# data_input_config: "DataInputConfig", # required
# },
# framework: "String",
# framework_version: "ModelPackageFrameworkVersion",
# nearest_model_name: "String",
# },
# ],
# supported_transform_instance_types: ["ml.m4.xlarge"], # accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# supported_realtime_inference_instance_types: ["ml.t2.medium"], # accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.12xlarge, ml.m5d.24xlarge, ml.c4.large, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.large, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.12xlarge, ml.r5.24xlarge, ml.r5d.large, ml.r5d.xlarge, ml.r5d.2xlarge, ml.r5d.4xlarge, ml.r5d.12xlarge, ml.r5d.24xlarge, ml.inf1.xlarge, ml.inf1.2xlarge, ml.inf1.6xlarge, ml.inf1.24xlarge, ml.c6i.large, ml.c6i.xlarge, ml.c6i.2xlarge, ml.c6i.4xlarge, ml.c6i.8xlarge, ml.c6i.12xlarge, ml.c6i.16xlarge, ml.c6i.24xlarge, ml.c6i.32xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.12xlarge, ml.g5.16xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.p4d.24xlarge, ml.c7g.large, ml.c7g.xlarge, ml.c7g.2xlarge, ml.c7g.4xlarge, ml.c7g.8xlarge, ml.c7g.12xlarge, ml.c7g.16xlarge, ml.m6g.large, ml.m6g.xlarge, ml.m6g.2xlarge, ml.m6g.4xlarge, ml.m6g.8xlarge, ml.m6g.12xlarge, ml.m6g.16xlarge, ml.m6gd.large, ml.m6gd.xlarge, ml.m6gd.2xlarge, ml.m6gd.4xlarge, ml.m6gd.8xlarge, ml.m6gd.12xlarge, ml.m6gd.16xlarge, ml.c6g.large, ml.c6g.xlarge, ml.c6g.2xlarge, ml.c6g.4xlarge, ml.c6g.8xlarge, ml.c6g.12xlarge, ml.c6g.16xlarge, ml.c6gd.large, ml.c6gd.xlarge, ml.c6gd.2xlarge, ml.c6gd.4xlarge, ml.c6gd.8xlarge, ml.c6gd.12xlarge, ml.c6gd.16xlarge, ml.c6gn.large, ml.c6gn.xlarge, ml.c6gn.2xlarge, ml.c6gn.4xlarge, ml.c6gn.8xlarge, ml.c6gn.12xlarge, ml.c6gn.16xlarge, ml.r6g.large, ml.r6g.xlarge, ml.r6g.2xlarge, ml.r6g.4xlarge, ml.r6g.8xlarge, ml.r6g.12xlarge, ml.r6g.16xlarge, ml.r6gd.large, ml.r6gd.xlarge, ml.r6gd.2xlarge, ml.r6gd.4xlarge, ml.r6gd.8xlarge, ml.r6gd.12xlarge, ml.r6gd.16xlarge, ml.p4de.24xlarge, ml.trn1.2xlarge, ml.trn1.32xlarge, ml.inf2.xlarge, ml.inf2.8xlarge, ml.inf2.24xlarge, ml.inf2.48xlarge
# supported_content_types: ["ContentType"],
# supported_response_mime_types: ["ResponseMIMEType"],
# },
# ],
# })
#
# @example Response structure
#
# resp.model_package_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelPackage AWS API Documentation
#
# @overload create_model_package(params = {})
# @param [Hash] params ({})
def create_model_package(params = {}, options = {})
req = build_request(:create_model_package, params)
req.send_request(options)
end
# Creates a model group. A model group contains a group of model
# versions.
#
# @option params [required, String] :model_package_group_name
# The name of the model group.
#
# @option params [String] :model_package_group_description
# A description for the model group.
#
# @option params [Array] :tags
# A list of key value pairs associated with the model group. For more
# information, see [Tagging Amazon Web Services resources][1] in the
# *Amazon Web Services General Reference Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @return [Types::CreateModelPackageGroupOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelPackageGroupOutput#model_package_group_arn #model_package_group_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_package_group({
# model_package_group_name: "EntityName", # required
# model_package_group_description: "EntityDescription",
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.model_package_group_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelPackageGroup AWS API Documentation
#
# @overload create_model_package_group(params = {})
# @param [Hash] params ({})
def create_model_package_group(params = {}, options = {})
req = build_request(:create_model_package_group, params)
req.send_request(options)
end
# Creates a definition for a job that monitors model quality and drift.
# For information about model monitor, see [Amazon SageMaker Model
# Monitor][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/model-monitor.html
#
# @option params [required, String] :job_definition_name
# The name of the monitoring job definition.
#
# @option params [Types::ModelQualityBaselineConfig] :model_quality_baseline_config
# Specifies the constraints and baselines for the monitoring job.
#
# @option params [required, Types::ModelQualityAppSpecification] :model_quality_app_specification
# The container that runs the monitoring job.
#
# @option params [required, Types::ModelQualityJobInput] :model_quality_job_input
# A list of the inputs that are monitored. Currently endpoints are
# supported.
#
# @option params [required, Types::MonitoringOutputConfig] :model_quality_job_output_config
# The output configuration for monitoring jobs.
#
# @option params [required, Types::MonitoringResources] :job_resources
# Identifies the resources to deploy for a monitoring job.
#
# @option params [Types::MonitoringNetworkConfig] :network_config
# Specifies the network configuration for the monitoring job.
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) of an IAM role that Amazon SageMaker
# can assume to perform tasks on your behalf.
#
# @option params [Types::MonitoringStoppingCondition] :stopping_condition
# A time limit for how long the monitoring job is allowed to run before
# stopping.
#
# @option params [Array] :tags
# (Optional) An array of key-value pairs. For more information, see
# [Using Cost Allocation Tags][1] in the *Amazon Web Services Billing
# and Cost Management User Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-whatURL
#
# @return [Types::CreateModelQualityJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateModelQualityJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_model_quality_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# model_quality_baseline_config: {
# baselining_job_name: "ProcessingJobName",
# constraints_resource: {
# s3_uri: "S3Uri",
# },
# },
# model_quality_app_specification: { # required
# image_uri: "ImageUri", # required
# container_entrypoint: ["ContainerEntrypointString"],
# container_arguments: ["ContainerArgument"],
# record_preprocessor_source_uri: "S3Uri",
# post_analytics_processor_source_uri: "S3Uri",
# problem_type: "BinaryClassification", # accepts BinaryClassification, MulticlassClassification, Regression
# environment: {
# "ProcessingEnvironmentKey" => "ProcessingEnvironmentValue",
# },
# },
# model_quality_job_input: { # required
# endpoint_input: {
# endpoint_name: "EndpointName", # required
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# batch_transform_input: {
# data_captured_destination_s3_uri: "DestinationS3Uri", # required
# dataset_format: { # required
# csv: {
# header: false,
# },
# json: {
# line: false,
# },
# parquet: {
# },
# },
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# ground_truth_s3_input: { # required
# s3_uri: "MonitoringS3Uri",
# },
# },
# model_quality_job_output_config: { # required
# monitoring_outputs: [ # required
# {
# s3_output: { # required
# s3_uri: "MonitoringS3Uri", # required
# local_path: "ProcessingLocalPath", # required
# s3_upload_mode: "Continuous", # accepts Continuous, EndOfJob
# },
# },
# ],
# kms_key_id: "KmsKeyId",
# },
# job_resources: { # required
# cluster_config: { # required
# instance_count: 1, # required
# instance_type: "ml.t3.medium", # required, accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# network_config: {
# enable_inter_container_traffic_encryption: false,
# enable_network_isolation: false,
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# role_arn: "RoleArn", # required
# stopping_condition: {
# max_runtime_in_seconds: 1, # required
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateModelQualityJobDefinition AWS API Documentation
#
# @overload create_model_quality_job_definition(params = {})
# @param [Hash] params ({})
def create_model_quality_job_definition(params = {}, options = {})
req = build_request(:create_model_quality_job_definition, params)
req.send_request(options)
end
# Creates a schedule that regularly starts Amazon SageMaker Processing
# Jobs to monitor the data captured for an Amazon SageMaker Endpoint.
#
# @option params [required, String] :monitoring_schedule_name
# The name of the monitoring schedule. The name must be unique within an
# Amazon Web Services Region within an Amazon Web Services account.
#
# @option params [required, Types::MonitoringScheduleConfig] :monitoring_schedule_config
# The configuration object that specifies the monitoring schedule and
# defines the monitoring job.
#
# @option params [Array] :tags
# (Optional) An array of key-value pairs. For more information, see
# [Using Cost Allocation Tags](
# https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-whatURL)
# in the *Amazon Web Services Billing and Cost Management User Guide*.
#
# @return [Types::CreateMonitoringScheduleResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateMonitoringScheduleResponse#monitoring_schedule_arn #monitoring_schedule_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_monitoring_schedule({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# monitoring_schedule_config: { # required
# schedule_config: {
# schedule_expression: "ScheduleExpression", # required
# },
# monitoring_job_definition: {
# baseline_config: {
# baselining_job_name: "ProcessingJobName",
# constraints_resource: {
# s3_uri: "S3Uri",
# },
# statistics_resource: {
# s3_uri: "S3Uri",
# },
# },
# monitoring_inputs: [ # required
# {
# endpoint_input: {
# endpoint_name: "EndpointName", # required
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# batch_transform_input: {
# data_captured_destination_s3_uri: "DestinationS3Uri", # required
# dataset_format: { # required
# csv: {
# header: false,
# },
# json: {
# line: false,
# },
# parquet: {
# },
# },
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# },
# ],
# monitoring_output_config: { # required
# monitoring_outputs: [ # required
# {
# s3_output: { # required
# s3_uri: "MonitoringS3Uri", # required
# local_path: "ProcessingLocalPath", # required
# s3_upload_mode: "Continuous", # accepts Continuous, EndOfJob
# },
# },
# ],
# kms_key_id: "KmsKeyId",
# },
# monitoring_resources: { # required
# cluster_config: { # required
# instance_count: 1, # required
# instance_type: "ml.t3.medium", # required, accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# monitoring_app_specification: { # required
# image_uri: "ImageUri", # required
# container_entrypoint: ["ContainerEntrypointString"],
# container_arguments: ["ContainerArgument"],
# record_preprocessor_source_uri: "S3Uri",
# post_analytics_processor_source_uri: "S3Uri",
# },
# stopping_condition: {
# max_runtime_in_seconds: 1, # required
# },
# environment: {
# "ProcessingEnvironmentKey" => "ProcessingEnvironmentValue",
# },
# network_config: {
# enable_inter_container_traffic_encryption: false,
# enable_network_isolation: false,
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# role_arn: "RoleArn", # required
# },
# monitoring_job_definition_name: "MonitoringJobDefinitionName",
# monitoring_type: "DataQuality", # accepts DataQuality, ModelQuality, ModelBias, ModelExplainability
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.monitoring_schedule_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateMonitoringSchedule AWS API Documentation
#
# @overload create_monitoring_schedule(params = {})
# @param [Hash] params ({})
def create_monitoring_schedule(params = {}, options = {})
req = build_request(:create_monitoring_schedule, params)
req.send_request(options)
end
# Creates an SageMaker notebook instance. A notebook instance is a
# machine learning (ML) compute instance running on a Jupyter notebook.
#
# In a `CreateNotebookInstance` request, specify the type of ML compute
# instance that you want to run. SageMaker launches the instance,
# installs common libraries that you can use to explore datasets for
# model training, and attaches an ML storage volume to the notebook
# instance.
#
# SageMaker also provides a set of example notebooks. Each notebook
# demonstrates how to use SageMaker with a specific algorithm or with a
# machine learning framework.
#
# After receiving the request, SageMaker does the following:
#
# 1. Creates a network interface in the SageMaker VPC.
#
# 2. (Option) If you specified `SubnetId`, SageMaker creates a network
# interface in your own VPC, which is inferred from the subnet ID
# that you provide in the input. When creating this network
# interface, SageMaker attaches the security group that you
# specified in the request to the network interface that it creates
# in your VPC.
#
# 3. Launches an EC2 instance of the type specified in the request in
# the SageMaker VPC. If you specified `SubnetId` of your VPC,
# SageMaker specifies both network interfaces when launching this
# instance. This enables inbound traffic from your own VPC to the
# notebook instance, assuming that the security groups allow it.
#
# After creating the notebook instance, SageMaker returns its Amazon
# Resource Name (ARN). You can't change the name of a notebook instance
# after you create it.
#
# After SageMaker creates the notebook instance, you can connect to the
# Jupyter server and work in Jupyter notebooks. For example, you can
# write code to explore a dataset that you can use for model training,
# train a model, host models by creating SageMaker endpoints, and
# validate hosted models.
#
# For more information, see [How It Works][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works.html
#
# @option params [required, String] :notebook_instance_name
# The name of the new notebook instance.
#
# @option params [required, String] :instance_type
# The type of ML compute instance to launch for the notebook instance.
#
# @option params [String] :subnet_id
# The ID of the subnet in a VPC to which you would like to have a
# connectivity from your ML compute instance.
#
# @option params [Array] :security_group_ids
# The VPC security group IDs, in the form sg-xxxxxxxx. The security
# groups must be for the same VPC as specified in the subnet.
#
# @option params [required, String] :role_arn
# When you send any requests to Amazon Web Services resources from the
# notebook instance, SageMaker assumes this role to perform tasks on
# your behalf. You must grant this role necessary permissions so
# SageMaker can perform these tasks. The policy must allow the SageMaker
# service principal (sagemaker.amazonaws.com) permissions to assume this
# role. For more information, see [SageMaker Roles][1].
#
# To be able to pass this role to SageMaker, the caller of this API must
# have the `iam:PassRole` permission.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html
#
# @option params [String] :kms_key_id
# The Amazon Resource Name (ARN) of a Amazon Web Services Key Management
# Service key that SageMaker uses to encrypt data on the storage volume
# attached to your notebook instance. The KMS key you provide must be
# enabled. For information, see [Enabling and Disabling Keys][1] in the
# *Amazon Web Services Key Management Service Developer Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/kms/latest/developerguide/enabling-keys.html
#
# @option params [Array] :tags
# An array of key-value pairs. You can use tags to categorize your
# Amazon Web Services resources in different ways, for example, by
# purpose, owner, or environment. For more information, see [Tagging
# Amazon Web Services Resources][1].
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @option params [String] :lifecycle_config_name
# The name of a lifecycle configuration to associate with the notebook
# instance. For information about lifestyle configurations, see [Step
# 2.1: (Optional) Customize a Notebook Instance][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/notebook-lifecycle-config.html
#
# @option params [String] :direct_internet_access
# Sets whether SageMaker provides internet access to the notebook
# instance. If you set this to `Disabled` this notebook instance is able
# to access resources only in your VPC, and is not be able to connect to
# SageMaker training and endpoint services unless you configure a NAT
# Gateway in your VPC.
#
# For more information, see [Notebook Instances Are Internet-Enabled by
# Default][1]. You can set the value of this parameter to `Disabled`
# only if you set a value for the `SubnetId` parameter.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/appendix-additional-considerations.html#appendix-notebook-and-internet-access
#
# @option params [Integer] :volume_size_in_gb
# The size, in GB, of the ML storage volume to attach to the notebook
# instance. The default value is 5 GB.
#
# @option params [Array] :accelerator_types
# A list of Elastic Inference (EI) instance types to associate with this
# notebook instance. Currently, only one instance type can be associated
# with a notebook instance. For more information, see [Using Elastic
# Inference in Amazon SageMaker][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/ei.html
#
# @option params [String] :default_code_repository
# A Git repository to associate with the notebook instance as its
# default code repository. This can be either the name of a Git
# repository stored as a resource in your account, or the URL of a Git
# repository in [Amazon Web Services CodeCommit][1] or in any other Git
# repository. When you open a notebook instance, it opens in the
# directory that contains this repository. For more information, see
# [Associating Git Repositories with SageMaker Notebook Instances][2].
#
#
#
# [1]: https://docs.aws.amazon.com/codecommit/latest/userguide/welcome.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/nbi-git-repo.html
#
# @option params [Array] :additional_code_repositories
# An array of up to three Git repositories to associate with the
# notebook instance. These can be either the names of Git repositories
# stored as resources in your account, or the URL of Git repositories in
# [Amazon Web Services CodeCommit][1] or in any other Git repository.
# These repositories are cloned at the same level as the default
# repository of your notebook instance. For more information, see
# [Associating Git Repositories with SageMaker Notebook Instances][2].
#
#
#
# [1]: https://docs.aws.amazon.com/codecommit/latest/userguide/welcome.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/nbi-git-repo.html
#
# @option params [String] :root_access
# Whether root access is enabled or disabled for users of the notebook
# instance. The default value is `Enabled`.
#
# Lifecycle configurations need root access to be able to set up a
# notebook instance. Because of this, lifecycle configurations
# associated with a notebook instance always run with root access even
# if you disable root access for users.
#
#
#
# @option params [String] :platform_identifier
# The platform identifier of the notebook instance runtime environment.
#
# @option params [Types::InstanceMetadataServiceConfiguration] :instance_metadata_service_configuration
# Information on the IMDS configuration of the notebook instance
#
# @return [Types::CreateNotebookInstanceOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateNotebookInstanceOutput#notebook_instance_arn #notebook_instance_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_notebook_instance({
# notebook_instance_name: "NotebookInstanceName", # required
# instance_type: "ml.t2.medium", # required, accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge
# subnet_id: "SubnetId",
# security_group_ids: ["SecurityGroupId"],
# role_arn: "RoleArn", # required
# kms_key_id: "KmsKeyId",
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# lifecycle_config_name: "NotebookInstanceLifecycleConfigName",
# direct_internet_access: "Enabled", # accepts Enabled, Disabled
# volume_size_in_gb: 1,
# accelerator_types: ["ml.eia1.medium"], # accepts ml.eia1.medium, ml.eia1.large, ml.eia1.xlarge, ml.eia2.medium, ml.eia2.large, ml.eia2.xlarge
# default_code_repository: "CodeRepositoryNameOrUrl",
# additional_code_repositories: ["CodeRepositoryNameOrUrl"],
# root_access: "Enabled", # accepts Enabled, Disabled
# platform_identifier: "PlatformIdentifier",
# instance_metadata_service_configuration: {
# minimum_instance_metadata_service_version: "MinimumInstanceMetadataServiceVersion", # required
# },
# })
#
# @example Response structure
#
# resp.notebook_instance_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateNotebookInstance AWS API Documentation
#
# @overload create_notebook_instance(params = {})
# @param [Hash] params ({})
def create_notebook_instance(params = {}, options = {})
req = build_request(:create_notebook_instance, params)
req.send_request(options)
end
# Creates a lifecycle configuration that you can associate with a
# notebook instance. A *lifecycle configuration* is a collection of
# shell scripts that run when you create or start a notebook instance.
#
# Each lifecycle configuration script has a limit of 16384 characters.
#
# The value of the `$PATH` environment variable that is available to
# both scripts is `/sbin:bin:/usr/sbin:/usr/bin`.
#
# View CloudWatch Logs for notebook instance lifecycle configurations in
# log group `/aws/sagemaker/NotebookInstances` in log stream
# `[notebook-instance-name]/[LifecycleConfigHook]`.
#
# Lifecycle configuration scripts cannot run for longer than 5 minutes.
# If a script runs for longer than 5 minutes, it fails and the notebook
# instance is not created or started.
#
# For information about notebook instance lifestyle configurations, see
# [Step 2.1: (Optional) Customize a Notebook Instance][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/notebook-lifecycle-config.html
#
# @option params [required, String] :notebook_instance_lifecycle_config_name
# The name of the lifecycle configuration.
#
# @option params [Array] :on_create
# A shell script that runs only once, when you create a notebook
# instance. The shell script must be a base64-encoded string.
#
# @option params [Array] :on_start
# A shell script that runs every time you start a notebook instance,
# including when you create the notebook instance. The shell script must
# be a base64-encoded string.
#
# @return [Types::CreateNotebookInstanceLifecycleConfigOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateNotebookInstanceLifecycleConfigOutput#notebook_instance_lifecycle_config_arn #notebook_instance_lifecycle_config_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_notebook_instance_lifecycle_config({
# notebook_instance_lifecycle_config_name: "NotebookInstanceLifecycleConfigName", # required
# on_create: [
# {
# content: "NotebookInstanceLifecycleConfigContent",
# },
# ],
# on_start: [
# {
# content: "NotebookInstanceLifecycleConfigContent",
# },
# ],
# })
#
# @example Response structure
#
# resp.notebook_instance_lifecycle_config_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateNotebookInstanceLifecycleConfig AWS API Documentation
#
# @overload create_notebook_instance_lifecycle_config(params = {})
# @param [Hash] params ({})
def create_notebook_instance_lifecycle_config(params = {}, options = {})
req = build_request(:create_notebook_instance_lifecycle_config, params)
req.send_request(options)
end
# Creates a pipeline using a JSON pipeline definition.
#
# @option params [required, String] :pipeline_name
# The name of the pipeline.
#
# @option params [String] :pipeline_display_name
# The display name of the pipeline.
#
# @option params [String] :pipeline_definition
# The JSON pipeline definition of the pipeline.
#
# @option params [Types::PipelineDefinitionS3Location] :pipeline_definition_s3_location
# The location of the pipeline definition stored in Amazon S3. If
# specified, SageMaker will retrieve the pipeline definition from this
# location.
#
# @option params [String] :pipeline_description
# A description of the pipeline.
#
# @option params [required, String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than one time.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) of the role used by the pipeline to
# access and create resources.
#
# @option params [Array] :tags
# A list of tags to apply to the created pipeline.
#
# @option params [Types::ParallelismConfiguration] :parallelism_configuration
# This is the configuration that controls the parallelism of the
# pipeline. If specified, it applies to all runs of this pipeline by
# default.
#
# @return [Types::CreatePipelineResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreatePipelineResponse#pipeline_arn #pipeline_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_pipeline({
# pipeline_name: "PipelineName", # required
# pipeline_display_name: "PipelineName",
# pipeline_definition: "PipelineDefinition",
# pipeline_definition_s3_location: {
# bucket: "BucketName", # required
# object_key: "Key", # required
# version_id: "VersionId",
# },
# pipeline_description: "PipelineDescription",
# client_request_token: "IdempotencyToken", # required
# role_arn: "RoleArn", # required
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# parallelism_configuration: {
# max_parallel_execution_steps: 1, # required
# },
# })
#
# @example Response structure
#
# resp.pipeline_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreatePipeline AWS API Documentation
#
# @overload create_pipeline(params = {})
# @param [Hash] params ({})
def create_pipeline(params = {}, options = {})
req = build_request(:create_pipeline, params)
req.send_request(options)
end
# Creates a URL for a specified UserProfile in a Domain. When accessed
# in a web browser, the user will be automatically signed in to Amazon
# SageMaker Studio, and granted access to all of the Apps and files
# associated with the Domain's Amazon Elastic File System (EFS) volume.
# This operation can only be called when the authentication mode equals
# IAM.
#
# The IAM role or user passed to this API defines the permissions to
# access the app. Once the presigned URL is created, no additional
# permission is required to access this URL. IAM authorization policies
# for this API are also enforced for every HTTP request and WebSocket
# frame that attempts to connect to the app.
#
# You can restrict access to this API and to the URL that it returns to
# a list of IP addresses, Amazon VPCs or Amazon VPC Endpoints that you
# specify. For more information, see [Connect to SageMaker Studio
# Through an Interface VPC Endpoint][1] .
#
# The URL that you get from a call to `CreatePresignedDomainUrl` has a
# default timeout of 5 minutes. You can configure this value using
# `ExpiresInSeconds`. If you try to use the URL after the timeout limit
# expires, you are directed to the Amazon Web Services console sign-in
# page.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/studio-interface-endpoint.html
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [required, String] :user_profile_name
# The name of the UserProfile to sign-in as.
#
# @option params [Integer] :session_expiration_duration_in_seconds
# The session expiration duration in seconds. This value defaults to
# 43200.
#
# @option params [Integer] :expires_in_seconds
# The number of seconds until the pre-signed URL expires. This value
# defaults to 300.
#
# @option params [String] :space_name
# The name of the space.
#
# @return [Types::CreatePresignedDomainUrlResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreatePresignedDomainUrlResponse#authorized_url #authorized_url} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_presigned_domain_url({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName", # required
# session_expiration_duration_in_seconds: 1,
# expires_in_seconds: 1,
# space_name: "SpaceName",
# })
#
# @example Response structure
#
# resp.authorized_url #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreatePresignedDomainUrl AWS API Documentation
#
# @overload create_presigned_domain_url(params = {})
# @param [Hash] params ({})
def create_presigned_domain_url(params = {}, options = {})
req = build_request(:create_presigned_domain_url, params)
req.send_request(options)
end
# Returns a URL that you can use to connect to the Jupyter server from a
# notebook instance. In the SageMaker console, when you choose `Open`
# next to a notebook instance, SageMaker opens a new tab showing the
# Jupyter server home page from the notebook instance. The console uses
# this API to get the URL and show the page.
#
# The IAM role or user used to call this API defines the permissions to
# access the notebook instance. Once the presigned URL is created, no
# additional permission is required to access this URL. IAM
# authorization policies for this API are also enforced for every HTTP
# request and WebSocket frame that attempts to connect to the notebook
# instance.
#
# You can restrict access to this API and to the URL that it returns to
# a list of IP addresses that you specify. Use the `NotIpAddress`
# condition operator and the `aws:SourceIP` condition context key to
# specify the list of IP addresses that you want to have access to the
# notebook instance. For more information, see [Limit Access to a
# Notebook Instance by IP Address][1].
#
# The URL that you get from a call to
# [CreatePresignedNotebookInstanceUrl][2] is valid only for 5 minutes.
# If you try to use the URL after the 5-minute limit expires, you are
# directed to the Amazon Web Services console sign-in page.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/security_iam_id-based-policy-examples.html#nbi-ip-filter
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreatePresignedNotebookInstanceUrl.html
#
# @option params [required, String] :notebook_instance_name
# The name of the notebook instance.
#
# @option params [Integer] :session_expiration_duration_in_seconds
# The duration of the session, in seconds. The default is 12 hours.
#
# @return [Types::CreatePresignedNotebookInstanceUrlOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreatePresignedNotebookInstanceUrlOutput#authorized_url #authorized_url} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_presigned_notebook_instance_url({
# notebook_instance_name: "NotebookInstanceName", # required
# session_expiration_duration_in_seconds: 1,
# })
#
# @example Response structure
#
# resp.authorized_url #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreatePresignedNotebookInstanceUrl AWS API Documentation
#
# @overload create_presigned_notebook_instance_url(params = {})
# @param [Hash] params ({})
def create_presigned_notebook_instance_url(params = {}, options = {})
req = build_request(:create_presigned_notebook_instance_url, params)
req.send_request(options)
end
# Creates a processing job.
#
# @option params [Array] :processing_inputs
# An array of inputs configuring the data to download into the
# processing container.
#
# @option params [Types::ProcessingOutputConfig] :processing_output_config
# Output configuration for the processing job.
#
# @option params [required, String] :processing_job_name
# The name of the processing job. The name must be unique within an
# Amazon Web Services Region in the Amazon Web Services account.
#
# @option params [required, Types::ProcessingResources] :processing_resources
# Identifies the resources, ML compute instances, and ML storage volumes
# to deploy for a processing job. In distributed training, you specify
# more than one instance.
#
# @option params [Types::ProcessingStoppingCondition] :stopping_condition
# The time limit for how long the processing job is allowed to run.
#
# @option params [required, Types::AppSpecification] :app_specification
# Configures the processing job to run a specified Docker container
# image.
#
# @option params [Hash] :environment
# The environment variables to set in the Docker container. Up to 100
# key and values entries in the map are supported.
#
# @option params [Types::NetworkConfig] :network_config
# Networking options for a processing job, such as whether to allow
# inbound and outbound network calls to and from processing containers,
# and the VPC subnets and security groups to use for VPC-enabled
# processing jobs.
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) of an IAM role that Amazon SageMaker
# can assume to perform tasks on your behalf.
#
# @option params [Array] :tags
# (Optional) An array of key-value pairs. For more information, see
# [Using Cost Allocation Tags][1] in the *Amazon Web Services Billing
# and Cost Management User Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-whatURL
#
# @option params [Types::ExperimentConfig] :experiment_config
# Associates a SageMaker job as a trial component with an experiment and
# trial. Specified when you call the following APIs:
#
# * [CreateProcessingJob][1]
#
# * [CreateTrainingJob][2]
#
# * [CreateTransformJob][3]
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateProcessingJob.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTrainingJob.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTransformJob.html
#
# @return [Types::CreateProcessingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateProcessingJobResponse#processing_job_arn #processing_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_processing_job({
# processing_inputs: [
# {
# input_name: "String", # required
# app_managed: false,
# s3_input: {
# s3_uri: "S3Uri", # required
# local_path: "ProcessingLocalPath",
# s3_data_type: "ManifestFile", # required, accepts ManifestFile, S3Prefix
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# s3_compression_type: "None", # accepts None, Gzip
# },
# dataset_definition: {
# athena_dataset_definition: {
# catalog: "AthenaCatalog", # required
# database: "AthenaDatabase", # required
# query_string: "AthenaQueryString", # required
# work_group: "AthenaWorkGroup",
# output_s3_uri: "S3Uri", # required
# kms_key_id: "KmsKeyId",
# output_format: "PARQUET", # required, accepts PARQUET, ORC, AVRO, JSON, TEXTFILE
# output_compression: "GZIP", # accepts GZIP, SNAPPY, ZLIB
# },
# redshift_dataset_definition: {
# cluster_id: "RedshiftClusterId", # required
# database: "RedshiftDatabase", # required
# db_user: "RedshiftUserName", # required
# query_string: "RedshiftQueryString", # required
# cluster_role_arn: "RoleArn", # required
# output_s3_uri: "S3Uri", # required
# kms_key_id: "KmsKeyId",
# output_format: "PARQUET", # required, accepts PARQUET, CSV
# output_compression: "None", # accepts None, GZIP, BZIP2, ZSTD, SNAPPY
# },
# local_path: "ProcessingLocalPath",
# data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# input_mode: "Pipe", # accepts Pipe, File
# },
# },
# ],
# processing_output_config: {
# outputs: [ # required
# {
# output_name: "String", # required
# s3_output: {
# s3_uri: "S3Uri", # required
# local_path: "ProcessingLocalPath", # required
# s3_upload_mode: "Continuous", # required, accepts Continuous, EndOfJob
# },
# feature_store_output: {
# feature_group_name: "FeatureGroupName", # required
# },
# app_managed: false,
# },
# ],
# kms_key_id: "KmsKeyId",
# },
# processing_job_name: "ProcessingJobName", # required
# processing_resources: { # required
# cluster_config: { # required
# instance_count: 1, # required
# instance_type: "ml.t3.medium", # required, accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# stopping_condition: {
# max_runtime_in_seconds: 1, # required
# },
# app_specification: { # required
# image_uri: "ImageUri", # required
# container_entrypoint: ["ContainerEntrypointString"],
# container_arguments: ["ContainerArgument"],
# },
# environment: {
# "ProcessingEnvironmentKey" => "ProcessingEnvironmentValue",
# },
# network_config: {
# enable_inter_container_traffic_encryption: false,
# enable_network_isolation: false,
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# role_arn: "RoleArn", # required
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# experiment_config: {
# experiment_name: "ExperimentEntityName",
# trial_name: "ExperimentEntityName",
# trial_component_display_name: "ExperimentEntityName",
# run_name: "ExperimentEntityName",
# },
# })
#
# @example Response structure
#
# resp.processing_job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateProcessingJob AWS API Documentation
#
# @overload create_processing_job(params = {})
# @param [Hash] params ({})
def create_processing_job(params = {}, options = {})
req = build_request(:create_processing_job, params)
req.send_request(options)
end
# Creates a machine learning (ML) project that can contain one or more
# templates that set up an ML pipeline from training to deploying an
# approved model.
#
# @option params [required, String] :project_name
# The name of the project.
#
# @option params [String] :project_description
# A description for the project.
#
# @option params [required, Types::ServiceCatalogProvisioningDetails] :service_catalog_provisioning_details
# The product ID and provisioning artifact ID to provision a service
# catalog. The provisioning artifact ID will default to the latest
# provisioning artifact ID of the product, if you don't provide the
# provisioning artifact ID. For more information, see [What is Amazon
# Web Services Service Catalog][1].
#
#
#
# [1]: https://docs.aws.amazon.com/servicecatalog/latest/adminguide/introduction.html
#
# @option params [Array] :tags
# An array of key-value pairs that you want to use to organize and track
# your Amazon Web Services resource costs. For more information, see
# [Tagging Amazon Web Services resources][1] in the *Amazon Web Services
# General Reference Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @return [Types::CreateProjectOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateProjectOutput#project_arn #project_arn} => String
# * {Types::CreateProjectOutput#project_id #project_id} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_project({
# project_name: "ProjectEntityName", # required
# project_description: "EntityDescription",
# service_catalog_provisioning_details: { # required
# product_id: "ServiceCatalogEntityId", # required
# provisioning_artifact_id: "ServiceCatalogEntityId",
# path_id: "ServiceCatalogEntityId",
# provisioning_parameters: [
# {
# key: "ProvisioningParameterKey",
# value: "ProvisioningParameterValue",
# },
# ],
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.project_arn #=> String
# resp.project_id #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateProject AWS API Documentation
#
# @overload create_project(params = {})
# @param [Hash] params ({})
def create_project(params = {}, options = {})
req = build_request(:create_project, params)
req.send_request(options)
end
# Creates a space used for real time collaboration in a Domain.
#
# @option params [required, String] :domain_id
# The ID of the associated Domain.
#
# @option params [required, String] :space_name
# The name of the space.
#
# @option params [Array] :tags
# Tags to associated with the space. Each tag consists of a key and an
# optional value. Tag keys must be unique for each resource. Tags are
# searchable using the `Search` API.
#
# @option params [Types::SpaceSettings] :space_settings
# A collection of space settings.
#
# @return [Types::CreateSpaceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateSpaceResponse#space_arn #space_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_space({
# domain_id: "DomainId", # required
# space_name: "SpaceName", # required
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# space_settings: {
# jupyter_server_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# code_repositories: [
# {
# repository_url: "RepositoryUrl", # required
# },
# ],
# },
# kernel_gateway_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# },
# },
# })
#
# @example Response structure
#
# resp.space_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateSpace AWS API Documentation
#
# @overload create_space(params = {})
# @param [Hash] params ({})
def create_space(params = {}, options = {})
req = build_request(:create_space, params)
req.send_request(options)
end
# Creates a new Studio Lifecycle Configuration.
#
# @option params [required, String] :studio_lifecycle_config_name
# The name of the Studio Lifecycle Configuration to create.
#
# @option params [required, String] :studio_lifecycle_config_content
# The content of your Studio Lifecycle Configuration script. This
# content must be base64 encoded.
#
# @option params [required, String] :studio_lifecycle_config_app_type
# The App type that the Lifecycle Configuration is attached to.
#
# @option params [Array] :tags
# Tags to be associated with the Lifecycle Configuration. Each tag
# consists of a key and an optional value. Tag keys must be unique per
# resource. Tags are searchable using the Search API.
#
# @return [Types::CreateStudioLifecycleConfigResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateStudioLifecycleConfigResponse#studio_lifecycle_config_arn #studio_lifecycle_config_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_studio_lifecycle_config({
# studio_lifecycle_config_name: "StudioLifecycleConfigName", # required
# studio_lifecycle_config_content: "StudioLifecycleConfigContent", # required
# studio_lifecycle_config_app_type: "JupyterServer", # required, accepts JupyterServer, KernelGateway
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.studio_lifecycle_config_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateStudioLifecycleConfig AWS API Documentation
#
# @overload create_studio_lifecycle_config(params = {})
# @param [Hash] params ({})
def create_studio_lifecycle_config(params = {}, options = {})
req = build_request(:create_studio_lifecycle_config, params)
req.send_request(options)
end
# Starts a model training job. After training completes, SageMaker saves
# the resulting model artifacts to an Amazon S3 location that you
# specify.
#
# If you choose to host your model using SageMaker hosting services, you
# can use the resulting model artifacts as part of the model. You can
# also use the artifacts in a machine learning service other than
# SageMaker, provided that you know how to use them for inference.
#
# In the request body, you provide the following:
#
# * `AlgorithmSpecification` - Identifies the training algorithm to use.
#
# * `HyperParameters` - Specify these algorithm-specific parameters to
# enable the estimation of model parameters during training.
# Hyperparameters can be tuned to optimize this learning process. For
# a list of hyperparameters for each training algorithm provided by
# SageMaker, see [Algorithms][1].
#
# Do not include any security-sensitive information including account
# access IDs, secrets or tokens in any hyperparameter field. If the
# use of security-sensitive credentials are detected, SageMaker will
# reject your training job request and return an exception error.
#
# * `InputDataConfig` - Describes the input required by the training job
# and the Amazon S3, EFS, or FSx location where it is stored.
#
# * `OutputDataConfig` - Identifies the Amazon S3 bucket where you want
# SageMaker to save the results of model training.
#
# * `ResourceConfig` - Identifies the resources, ML compute instances,
# and ML storage volumes to deploy for model training. In distributed
# training, you specify more than one instance.
#
# * `EnableManagedSpotTraining` - Optimize the cost of training machine
# learning models by up to 80% by using Amazon EC2 Spot instances. For
# more information, see [Managed Spot Training][2].
#
# * `RoleArn` - The Amazon Resource Name (ARN) that SageMaker assumes to
# perform tasks on your behalf during model training. You must grant
# this role the necessary permissions so that SageMaker can
# successfully complete model training.
#
# * `StoppingCondition` - To help cap training costs, use
# `MaxRuntimeInSeconds` to set a time limit for training. Use
# `MaxWaitTimeInSeconds` to specify how long a managed spot training
# job has to complete.
#
# * `Environment` - The environment variables to set in the Docker
# container.
#
# * `RetryStrategy` - The number of times to retry the job when the job
# fails due to an `InternalServerError`.
#
# For more information about SageMaker, see [How It Works][3].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/algos.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/model-managed-spot-training.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works.html
#
# @option params [required, String] :training_job_name
# The name of the training job. The name must be unique within an Amazon
# Web Services Region in an Amazon Web Services account.
#
# @option params [Hash] :hyper_parameters
# Algorithm-specific parameters that influence the quality of the model.
# You set hyperparameters before you start the learning process. For a
# list of hyperparameters for each training algorithm provided by
# SageMaker, see [Algorithms][1].
#
# You can specify a maximum of 100 hyperparameters. Each hyperparameter
# is a key-value pair. Each key and value is limited to 256 characters,
# as specified by the `Length Constraint`.
#
# Do not include any security-sensitive information including account
# access IDs, secrets or tokens in any hyperparameter field. If the use
# of security-sensitive credentials are detected, SageMaker will reject
# your training job request and return an exception error.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/algos.html
#
# @option params [required, Types::AlgorithmSpecification] :algorithm_specification
# The registry path of the Docker image that contains the training
# algorithm and algorithm-specific metadata, including the input mode.
# For more information about algorithms provided by SageMaker, see
# [Algorithms][1]. For information about providing your own algorithms,
# see [Using Your Own Algorithms with Amazon SageMaker][2].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/algos.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms.html
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) of an IAM role that SageMaker can
# assume to perform tasks on your behalf.
#
# During model training, SageMaker needs your permission to read input
# data from an S3 bucket, download a Docker image that contains training
# code, write model artifacts to an S3 bucket, write logs to Amazon
# CloudWatch Logs, and publish metrics to Amazon CloudWatch. You grant
# permissions for all of these tasks to an IAM role. For more
# information, see [SageMaker Roles][1].
#
# To be able to pass this role to SageMaker, the caller of this API must
# have the `iam:PassRole` permission.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html
#
# @option params [Array] :input_data_config
# An array of `Channel` objects. Each channel is a named input source.
# `InputDataConfig` describes the input data and its location.
#
# Algorithms can accept input data from one or more channels. For
# example, an algorithm might have two channels of input data,
# `training_data` and `validation_data`. The configuration for each
# channel provides the S3, EFS, or FSx location where the input data is
# stored. It also provides information about the stored data: the MIME
# type, compression method, and whether the data is wrapped in RecordIO
# format.
#
# Depending on the input mode that the algorithm supports, SageMaker
# either copies input data files from an S3 bucket to a local directory
# in the Docker container, or makes it available as input streams. For
# example, if you specify an EFS location, input data files are
# available as input streams. They do not need to be downloaded.
#
# @option params [required, Types::OutputDataConfig] :output_data_config
# Specifies the path to the S3 location where you want to store model
# artifacts. SageMaker creates subfolders for the artifacts.
#
# @option params [required, Types::ResourceConfig] :resource_config
# The resources, including the ML compute instances and ML storage
# volumes, to use for model training.
#
# ML storage volumes store model artifacts and incremental states.
# Training algorithms might also use ML storage volumes for scratch
# space. If you want SageMaker to use the ML storage volume to store the
# training data, choose `File` as the `TrainingInputMode` in the
# algorithm specification. For distributed training algorithms, specify
# an instance count greater than 1.
#
# @option params [Types::VpcConfig] :vpc_config
# A [VpcConfig][1] object that specifies the VPC that you want your
# training job to connect to. Control access to and from your training
# container by configuring the VPC. For more information, see [Protect
# Training Jobs by Using an Amazon Virtual Private Cloud][2].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_VpcConfig.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/train-vpc.html
#
# @option params [required, Types::StoppingCondition] :stopping_condition
# Specifies a limit to how long a model training job can run. It also
# specifies how long a managed Spot training job has to complete. When
# the job reaches the time limit, SageMaker ends the training job. Use
# this API to cap model training costs.
#
# To stop a job, SageMaker sends the algorithm the `SIGTERM` signal,
# which delays job termination for 120 seconds. Algorithms can use this
# 120-second window to save the model artifacts, so the results of
# training are not lost.
#
# @option params [Array] :tags
# An array of key-value pairs. You can use tags to categorize your
# Amazon Web Services resources in different ways, for example, by
# purpose, owner, or environment. For more information, see [Tagging
# Amazon Web Services Resources][1].
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
#
# @option params [Boolean] :enable_network_isolation
# Isolates the training container. No inbound or outbound network calls
# can be made, except for calls between peers within a training cluster
# for distributed training. If you enable network isolation for training
# jobs that are configured to use a VPC, SageMaker downloads and uploads
# customer data and model artifacts through the specified VPC, but the
# training container does not have network access.
#
# @option params [Boolean] :enable_inter_container_traffic_encryption
# To encrypt all communications between ML compute instances in
# distributed training, choose `True`. Encryption provides greater
# security for distributed training, but training might take longer. How
# long it takes depends on the amount of communication between compute
# instances, especially if you use a deep learning algorithm in
# distributed training. For more information, see [Protect
# Communications Between ML Compute Instances in a Distributed Training
# Job][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/train-encrypt.html
#
# @option params [Boolean] :enable_managed_spot_training
# To train models using managed spot training, choose `True`. Managed
# spot training provides a fully managed and scalable infrastructure for
# training machine learning models. this option is useful when training
# jobs can be interrupted and when there is flexibility when the
# training job is run.
#
# The complete and intermediate results of jobs are stored in an Amazon
# S3 bucket, and can be used as a starting point to train models
# incrementally. Amazon SageMaker provides metrics and logs in
# CloudWatch. They can be used to see when managed spot training jobs
# are running, interrupted, resumed, or completed.
#
# @option params [Types::CheckpointConfig] :checkpoint_config
# Contains information about the output location for managed spot
# training checkpoint data.
#
# @option params [Types::DebugHookConfig] :debug_hook_config
# Configuration information for the Amazon SageMaker Debugger hook
# parameters, metric and tensor collections, and storage paths. To learn
# more about how to configure the `DebugHookConfig` parameter, see [Use
# the SageMaker and Debugger Configuration API Operations to Create,
# Update, and Debug Your Training Job][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/debugger-createtrainingjob-api.html
#
# @option params [Array] :debug_rule_configurations
# Configuration information for Amazon SageMaker Debugger rules for
# debugging output tensors.
#
# @option params [Types::TensorBoardOutputConfig] :tensor_board_output_config
# Configuration of storage locations for the Amazon SageMaker Debugger
# TensorBoard output data.
#
# @option params [Types::ExperimentConfig] :experiment_config
# Associates a SageMaker job as a trial component with an experiment and
# trial. Specified when you call the following APIs:
#
# * [CreateProcessingJob][1]
#
# * [CreateTrainingJob][2]
#
# * [CreateTransformJob][3]
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateProcessingJob.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTrainingJob.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTransformJob.html
#
# @option params [Types::ProfilerConfig] :profiler_config
# Configuration information for Amazon SageMaker Debugger system
# monitoring, framework profiling, and storage paths.
#
# @option params [Array] :profiler_rule_configurations
# Configuration information for Amazon SageMaker Debugger rules for
# profiling system and framework metrics.
#
# @option params [Hash] :environment
# The environment variables to set in the Docker container.
#
# @option params [Types::RetryStrategy] :retry_strategy
# The number of times to retry the job when the job fails due to an
# `InternalServerError`.
#
# @return [Types::CreateTrainingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateTrainingJobResponse#training_job_arn #training_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_training_job({
# training_job_name: "TrainingJobName", # required
# hyper_parameters: {
# "HyperParameterKey" => "HyperParameterValue",
# },
# algorithm_specification: { # required
# training_image: "AlgorithmImage",
# algorithm_name: "ArnOrName",
# training_input_mode: "Pipe", # required, accepts Pipe, File, FastFile
# metric_definitions: [
# {
# name: "MetricName", # required
# regex: "MetricRegex", # required
# },
# ],
# enable_sage_maker_metrics_time_series: false,
# container_entrypoint: ["TrainingContainerEntrypointString"],
# container_arguments: ["TrainingContainerArgument"],
# training_image_config: {
# training_repository_access_mode: "Platform", # required, accepts Platform, Vpc
# training_repository_auth_config: {
# training_repository_credentials_provider_arn: "TrainingRepositoryCredentialsProviderArn", # required
# },
# },
# },
# role_arn: "RoleArn", # required
# input_data_config: [
# {
# channel_name: "ChannelName", # required
# data_source: { # required
# s3_data_source: {
# s3_data_type: "ManifestFile", # required, accepts ManifestFile, S3Prefix, AugmentedManifestFile
# s3_uri: "S3Uri", # required
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# attribute_names: ["AttributeName"],
# instance_group_names: ["InstanceGroupName"],
# },
# file_system_data_source: {
# file_system_id: "FileSystemId", # required
# file_system_access_mode: "rw", # required, accepts rw, ro
# file_system_type: "EFS", # required, accepts EFS, FSxLustre
# directory_path: "DirectoryPath", # required
# },
# },
# content_type: "ContentType",
# compression_type: "None", # accepts None, Gzip
# record_wrapper_type: "None", # accepts None, RecordIO
# input_mode: "Pipe", # accepts Pipe, File, FastFile
# shuffle_config: {
# seed: 1, # required
# },
# },
# ],
# output_data_config: { # required
# kms_key_id: "KmsKeyId",
# s3_output_path: "S3Uri", # required
# },
# resource_config: { # required
# instance_type: "ml.m4.xlarge", # accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.p4d.24xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5n.xlarge, ml.c5n.2xlarge, ml.c5n.4xlarge, ml.c5n.9xlarge, ml.c5n.18xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.trn1.2xlarge, ml.trn1.32xlarge
# instance_count: 1,
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# instance_groups: [
# {
# instance_type: "ml.m4.xlarge", # required, accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.p4d.24xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5n.xlarge, ml.c5n.2xlarge, ml.c5n.4xlarge, ml.c5n.9xlarge, ml.c5n.18xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.trn1.2xlarge, ml.trn1.32xlarge
# instance_count: 1, # required
# instance_group_name: "InstanceGroupName", # required
# },
# ],
# keep_alive_period_in_seconds: 1,
# },
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# stopping_condition: { # required
# max_runtime_in_seconds: 1,
# max_wait_time_in_seconds: 1,
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# enable_network_isolation: false,
# enable_inter_container_traffic_encryption: false,
# enable_managed_spot_training: false,
# checkpoint_config: {
# s3_uri: "S3Uri", # required
# local_path: "DirectoryPath",
# },
# debug_hook_config: {
# local_path: "DirectoryPath",
# s3_output_path: "S3Uri", # required
# hook_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# collection_configurations: [
# {
# collection_name: "CollectionName",
# collection_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# },
# ],
# },
# debug_rule_configurations: [
# {
# rule_configuration_name: "RuleConfigurationName", # required
# local_path: "DirectoryPath",
# s3_output_path: "S3Uri",
# rule_evaluator_image: "AlgorithmImage", # required
# instance_type: "ml.t3.medium", # accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1,
# rule_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# },
# ],
# tensor_board_output_config: {
# local_path: "DirectoryPath",
# s3_output_path: "S3Uri", # required
# },
# experiment_config: {
# experiment_name: "ExperimentEntityName",
# trial_name: "ExperimentEntityName",
# trial_component_display_name: "ExperimentEntityName",
# run_name: "ExperimentEntityName",
# },
# profiler_config: {
# s3_output_path: "S3Uri",
# profiling_interval_in_milliseconds: 1,
# profiling_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# disable_profiler: false,
# },
# profiler_rule_configurations: [
# {
# rule_configuration_name: "RuleConfigurationName", # required
# local_path: "DirectoryPath",
# s3_output_path: "S3Uri",
# rule_evaluator_image: "AlgorithmImage", # required
# instance_type: "ml.t3.medium", # accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1,
# rule_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# },
# ],
# environment: {
# "TrainingEnvironmentKey" => "TrainingEnvironmentValue",
# },
# retry_strategy: {
# maximum_retry_attempts: 1, # required
# },
# })
#
# @example Response structure
#
# resp.training_job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateTrainingJob AWS API Documentation
#
# @overload create_training_job(params = {})
# @param [Hash] params ({})
def create_training_job(params = {}, options = {})
req = build_request(:create_training_job, params)
req.send_request(options)
end
# Starts a transform job. A transform job uses a trained model to get
# inferences on a dataset and saves these results to an Amazon S3
# location that you specify.
#
# To perform batch transformations, you create a transform job and use
# the data that you have readily available.
#
# In the request body, you provide the following:
#
# * `TransformJobName` - Identifies the transform job. The name must be
# unique within an Amazon Web Services Region in an Amazon Web
# Services account.
#
# * `ModelName` - Identifies the model to use. `ModelName` must be the
# name of an existing Amazon SageMaker model in the same Amazon Web
# Services Region and Amazon Web Services account. For information on
# creating a model, see [CreateModel][1].
#
# * `TransformInput` - Describes the dataset to be transformed and the
# Amazon S3 location where it is stored.
#
# * `TransformOutput` - Identifies the Amazon S3 location where you want
# Amazon SageMaker to save the results from the transform job.
#
# * `TransformResources` - Identifies the ML compute instances for the
# transform job.
#
# For more information about how batch transformation works, see [Batch
# Transform][2].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateModel.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/batch-transform.html
#
# @option params [required, String] :transform_job_name
# The name of the transform job. The name must be unique within an
# Amazon Web Services Region in an Amazon Web Services account.
#
# @option params [required, String] :model_name
# The name of the model that you want to use for the transform job.
# `ModelName` must be the name of an existing Amazon SageMaker model
# within an Amazon Web Services Region in an Amazon Web Services
# account.
#
# @option params [Integer] :max_concurrent_transforms
# The maximum number of parallel requests that can be sent to each
# instance in a transform job. If `MaxConcurrentTransforms` is set to
# `0` or left unset, Amazon SageMaker checks the optional
# execution-parameters to determine the settings for your chosen
# algorithm. If the execution-parameters endpoint is not enabled, the
# default value is `1`. For more information on execution-parameters,
# see [How Containers Serve Requests][1]. For built-in algorithms, you
# don't need to set a value for `MaxConcurrentTransforms`.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms-batch-code.html#your-algorithms-batch-code-how-containe-serves-requests
#
# @option params [Types::ModelClientConfig] :model_client_config
# Configures the timeout and maximum number of retries for processing a
# transform job invocation.
#
# @option params [Integer] :max_payload_in_mb
# The maximum allowed size of the payload, in MB. A *payload* is the
# data portion of a record (without metadata). The value in
# `MaxPayloadInMB` must be greater than, or equal to, the size of a
# single record. To estimate the size of a record in MB, divide the size
# of your dataset by the number of records. To ensure that the records
# fit within the maximum payload size, we recommend using a slightly
# larger value. The default value is `6` MB.
#
# The value of `MaxPayloadInMB` cannot be greater than 100 MB. If you
# specify the `MaxConcurrentTransforms` parameter, the value of
# `(MaxConcurrentTransforms * MaxPayloadInMB)` also cannot exceed 100
# MB.
#
# For cases where the payload might be arbitrarily large and is
# transmitted using HTTP chunked encoding, set the value to `0`. This
# feature works only in supported algorithms. Currently, Amazon
# SageMaker built-in algorithms do not support HTTP chunked encoding.
#
# @option params [String] :batch_strategy
# Specifies the number of records to include in a mini-batch for an HTTP
# inference request. A *record* ** is a single unit of input data that
# inference can be made on. For example, a single line in a CSV file is
# a record.
#
# To enable the batch strategy, you must set the `SplitType` property to
# `Line`, `RecordIO`, or `TFRecord`.
#
# To use only one record when making an HTTP invocation request to a
# container, set `BatchStrategy` to `SingleRecord` and `SplitType` to
# `Line`.
#
# To fit as many records in a mini-batch as can fit within the
# `MaxPayloadInMB` limit, set `BatchStrategy` to `MultiRecord` and
# `SplitType` to `Line`.
#
# @option params [Hash] :environment
# The environment variables to set in the Docker container. We support
# up to 16 key and values entries in the map.
#
# @option params [required, Types::TransformInput] :transform_input
# Describes the input source and the way the transform job consumes it.
#
# @option params [required, Types::TransformOutput] :transform_output
# Describes the results of the transform job.
#
# @option params [Types::BatchDataCaptureConfig] :data_capture_config
# Configuration to control how SageMaker captures inference data.
#
# @option params [required, Types::TransformResources] :transform_resources
# Describes the resources, including ML instance types and ML instance
# count, to use for the transform job.
#
# @option params [Types::DataProcessing] :data_processing
# The data structure used to specify the data to be used for inference
# in a batch transform job and to associate the data that is relevant to
# the prediction results in the output. The input filter provided allows
# you to exclude input data that is not needed for inference in a batch
# transform job. The output filter provided allows you to include input
# data relevant to interpreting the predictions in the output from the
# job. For more information, see [Associate Prediction Results with
# their Corresponding Input Records][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/batch-transform-data-processing.html
#
# @option params [Array] :tags
# (Optional) An array of key-value pairs. For more information, see
# [Using Cost Allocation Tags][1] in the *Amazon Web Services Billing
# and Cost Management User Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-what
#
# @option params [Types::ExperimentConfig] :experiment_config
# Associates a SageMaker job as a trial component with an experiment and
# trial. Specified when you call the following APIs:
#
# * [CreateProcessingJob][1]
#
# * [CreateTrainingJob][2]
#
# * [CreateTransformJob][3]
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateProcessingJob.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTrainingJob.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTransformJob.html
#
# @return [Types::CreateTransformJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateTransformJobResponse#transform_job_arn #transform_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_transform_job({
# transform_job_name: "TransformJobName", # required
# model_name: "ModelName", # required
# max_concurrent_transforms: 1,
# model_client_config: {
# invocations_timeout_in_seconds: 1,
# invocations_max_retries: 1,
# },
# max_payload_in_mb: 1,
# batch_strategy: "MultiRecord", # accepts MultiRecord, SingleRecord
# environment: {
# "TransformEnvironmentKey" => "TransformEnvironmentValue",
# },
# transform_input: { # required
# data_source: { # required
# s3_data_source: { # required
# s3_data_type: "ManifestFile", # required, accepts ManifestFile, S3Prefix, AugmentedManifestFile
# s3_uri: "S3Uri", # required
# },
# },
# content_type: "ContentType",
# compression_type: "None", # accepts None, Gzip
# split_type: "None", # accepts None, Line, RecordIO, TFRecord
# },
# transform_output: { # required
# s3_output_path: "S3Uri", # required
# accept: "Accept",
# assemble_with: "None", # accepts None, Line
# kms_key_id: "KmsKeyId",
# },
# data_capture_config: {
# destination_s3_uri: "S3Uri", # required
# kms_key_id: "KmsKeyId",
# generate_inference_id: false,
# },
# transform_resources: { # required
# instance_type: "ml.m4.xlarge", # required, accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# instance_count: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# data_processing: {
# input_filter: "JsonPath",
# output_filter: "JsonPath",
# join_source: "Input", # accepts Input, None
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# experiment_config: {
# experiment_name: "ExperimentEntityName",
# trial_name: "ExperimentEntityName",
# trial_component_display_name: "ExperimentEntityName",
# run_name: "ExperimentEntityName",
# },
# })
#
# @example Response structure
#
# resp.transform_job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateTransformJob AWS API Documentation
#
# @overload create_transform_job(params = {})
# @param [Hash] params ({})
def create_transform_job(params = {}, options = {})
req = build_request(:create_transform_job, params)
req.send_request(options)
end
# Creates an SageMaker *trial*. A trial is a set of steps called *trial
# components* that produce a machine learning model. A trial is part of
# a single SageMaker *experiment*.
#
# When you use SageMaker Studio or the SageMaker Python SDK, all
# experiments, trials, and trial components are automatically tracked,
# logged, and indexed. When you use the Amazon Web Services SDK for
# Python (Boto), you must use the logging APIs provided by the SDK.
#
# You can add tags to a trial and then use the [Search][1] API to search
# for the tags.
#
# To get a list of all your trials, call the [ListTrials][2] API. To
# view a trial's properties, call the [DescribeTrial][3] API. To create
# a trial component, call the [CreateTrialComponent][4] API.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_Search.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ListTrials.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeTrial.html
# [4]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTrialComponent.html
#
# @option params [required, String] :trial_name
# The name of the trial. The name must be unique in your Amazon Web
# Services account and is not case-sensitive.
#
# @option params [String] :display_name
# The name of the trial as displayed. The name doesn't need to be
# unique. If `DisplayName` isn't specified, `TrialName` is displayed.
#
# @option params [required, String] :experiment_name
# The name of the experiment to associate the trial with.
#
# @option params [Types::MetadataProperties] :metadata_properties
# Metadata properties of the tracking entity, trial, or trial component.
#
# @option params [Array] :tags
# A list of tags to associate with the trial. You can use [Search][1]
# API to search on the tags.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_Search.html
#
# @return [Types::CreateTrialResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateTrialResponse#trial_arn #trial_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_trial({
# trial_name: "ExperimentEntityName", # required
# display_name: "ExperimentEntityName",
# experiment_name: "ExperimentEntityName", # required
# metadata_properties: {
# commit_id: "MetadataPropertyValue",
# repository: "MetadataPropertyValue",
# generated_by: "MetadataPropertyValue",
# project_id: "MetadataPropertyValue",
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.trial_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateTrial AWS API Documentation
#
# @overload create_trial(params = {})
# @param [Hash] params ({})
def create_trial(params = {}, options = {})
req = build_request(:create_trial, params)
req.send_request(options)
end
# Creates a *trial component*, which is a stage of a machine learning
# *trial*. A trial is composed of one or more trial components. A trial
# component can be used in multiple trials.
#
# Trial components include pre-processing jobs, training jobs, and batch
# transform jobs.
#
# When you use SageMaker Studio or the SageMaker Python SDK, all
# experiments, trials, and trial components are automatically tracked,
# logged, and indexed. When you use the Amazon Web Services SDK for
# Python (Boto), you must use the logging APIs provided by the SDK.
#
# You can add tags to a trial component and then use the [Search][1] API
# to search for the tags.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_Search.html
#
# @option params [required, String] :trial_component_name
# The name of the component. The name must be unique in your Amazon Web
# Services account and is not case-sensitive.
#
# @option params [String] :display_name
# The name of the component as displayed. The name doesn't need to be
# unique. If `DisplayName` isn't specified, `TrialComponentName` is
# displayed.
#
# @option params [Types::TrialComponentStatus] :status
# The status of the component. States include:
#
# * InProgress
#
# * Completed
#
# * Failed
#
# @option params [Time,DateTime,Date,Integer,String] :start_time
# When the component started.
#
# @option params [Time,DateTime,Date,Integer,String] :end_time
# When the component ended.
#
# @option params [Hash] :parameters
# The hyperparameters for the component.
#
# @option params [Hash] :input_artifacts
# The input artifacts for the component. Examples of input artifacts are
# datasets, algorithms, hyperparameters, source code, and instance
# types.
#
# @option params [Hash] :output_artifacts
# The output artifacts for the component. Examples of output artifacts
# are metrics, snapshots, logs, and images.
#
# @option params [Types::MetadataProperties] :metadata_properties
# Metadata properties of the tracking entity, trial, or trial component.
#
# @option params [Array] :tags
# A list of tags to associate with the component. You can use
# [Search][1] API to search on the tags.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_Search.html
#
# @return [Types::CreateTrialComponentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateTrialComponentResponse#trial_component_arn #trial_component_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_trial_component({
# trial_component_name: "ExperimentEntityName", # required
# display_name: "ExperimentEntityName",
# status: {
# primary_status: "InProgress", # accepts InProgress, Completed, Failed, Stopping, Stopped
# message: "TrialComponentStatusMessage",
# },
# start_time: Time.now,
# end_time: Time.now,
# parameters: {
# "TrialComponentKey256" => {
# string_value: "StringParameterValue",
# number_value: 1.0,
# },
# },
# input_artifacts: {
# "TrialComponentKey64" => {
# media_type: "MediaType",
# value: "TrialComponentArtifactValue", # required
# },
# },
# output_artifacts: {
# "TrialComponentKey64" => {
# media_type: "MediaType",
# value: "TrialComponentArtifactValue", # required
# },
# },
# metadata_properties: {
# commit_id: "MetadataPropertyValue",
# repository: "MetadataPropertyValue",
# generated_by: "MetadataPropertyValue",
# project_id: "MetadataPropertyValue",
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.trial_component_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateTrialComponent AWS API Documentation
#
# @overload create_trial_component(params = {})
# @param [Hash] params ({})
def create_trial_component(params = {}, options = {})
req = build_request(:create_trial_component, params)
req.send_request(options)
end
# Creates a user profile. A user profile represents a single user within
# a domain, and is the main way to reference a "person" for the
# purposes of sharing, reporting, and other user-oriented features. This
# entity is created when a user onboards to Amazon SageMaker Studio. If
# an administrator invites a person by email or imports them from IAM
# Identity Center, a user profile is automatically created. A user
# profile is the primary holder of settings for an individual user and
# has a reference to the user's private Amazon Elastic File System
# (EFS) home directory.
#
# @option params [required, String] :domain_id
# The ID of the associated Domain.
#
# @option params [required, String] :user_profile_name
# A name for the UserProfile. This value is not case sensitive.
#
# @option params [String] :single_sign_on_user_identifier
# A specifier for the type of value specified in SingleSignOnUserValue.
# Currently, the only supported value is "UserName". If the Domain's
# AuthMode is IAM Identity Center, this field is required. If the
# Domain's AuthMode is not IAM Identity Center, this field cannot be
# specified.
#
# @option params [String] :single_sign_on_user_value
# The username of the associated Amazon Web Services Single Sign-On User
# for this UserProfile. If the Domain's AuthMode is IAM Identity
# Center, this field is required, and must match a valid username of a
# user in your directory. If the Domain's AuthMode is not IAM Identity
# Center, this field cannot be specified.
#
# @option params [Array] :tags
# Each tag consists of a key and an optional value. Tag keys must be
# unique per resource.
#
# Tags that you specify for the User Profile are also added to all Apps
# that the User Profile launches.
#
# @option params [Types::UserSettings] :user_settings
# A collection of settings.
#
# @return [Types::CreateUserProfileResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateUserProfileResponse#user_profile_arn #user_profile_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_user_profile({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName", # required
# single_sign_on_user_identifier: "SingleSignOnUserIdentifier",
# single_sign_on_user_value: "String256",
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# user_settings: {
# execution_role: "RoleArn",
# security_groups: ["SecurityGroupId"],
# sharing_settings: {
# notebook_output_option: "Allowed", # accepts Allowed, Disabled
# s3_output_path: "S3Uri",
# s3_kms_key_id: "KmsKeyId",
# },
# jupyter_server_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# code_repositories: [
# {
# repository_url: "RepositoryUrl", # required
# },
# ],
# },
# kernel_gateway_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# },
# tensor_board_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# },
# r_studio_server_pro_app_settings: {
# access_status: "ENABLED", # accepts ENABLED, DISABLED
# user_group: "R_STUDIO_ADMIN", # accepts R_STUDIO_ADMIN, R_STUDIO_USER
# },
# r_session_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# },
# canvas_app_settings: {
# time_series_forecasting_settings: {
# status: "ENABLED", # accepts ENABLED, DISABLED
# amazon_forecast_role_arn: "RoleArn",
# },
# model_register_settings: {
# status: "ENABLED", # accepts ENABLED, DISABLED
# cross_account_model_register_role_arn: "RoleArn",
# },
# },
# },
# })
#
# @example Response structure
#
# resp.user_profile_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateUserProfile AWS API Documentation
#
# @overload create_user_profile(params = {})
# @param [Hash] params ({})
def create_user_profile(params = {}, options = {})
req = build_request(:create_user_profile, params)
req.send_request(options)
end
# Use this operation to create a workforce. This operation will return
# an error if a workforce already exists in the Amazon Web Services
# Region that you specify. You can only create one workforce in each
# Amazon Web Services Region per Amazon Web Services account.
#
# If you want to create a new workforce in an Amazon Web Services Region
# where a workforce already exists, use the [DeleteWorkforce][1] API
# operation to delete the existing workforce and then use
# `CreateWorkforce` to create a new workforce.
#
# To create a private workforce using Amazon Cognito, you must specify a
# Cognito user pool in `CognitoConfig`. You can also create an Amazon
# Cognito workforce using the Amazon SageMaker console. For more
# information, see [ Create a Private Workforce (Amazon Cognito)][2].
#
# To create a private workforce using your own OIDC Identity Provider
# (IdP), specify your IdP configuration in `OidcConfig`. Your OIDC IdP
# must support *groups* because groups are used by Ground Truth and
# Amazon A2I to create work teams. For more information, see [ Create a
# Private Workforce (OIDC IdP)][3].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DeleteWorkforce.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-workforce-create-private.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-workforce-create-private-oidc.html
#
# @option params [Types::CognitoConfig] :cognito_config
# Use this parameter to configure an Amazon Cognito private workforce. A
# single Cognito workforce is created using and corresponds to a single
# [ Amazon Cognito user pool][1].
#
# Do not use `OidcConfig` if you specify values for `CognitoConfig`.
#
#
#
# [1]: https://docs.aws.amazon.com/cognito/latest/developerguide/cognito-user-identity-pools.html
#
# @option params [Types::OidcConfig] :oidc_config
# Use this parameter to configure a private workforce using your own
# OIDC Identity Provider.
#
# Do not use `CognitoConfig` if you specify values for `OidcConfig`.
#
# @option params [Types::SourceIpConfig] :source_ip_config
# A list of IP address ranges ([CIDRs][1]). Used to create an allow list
# of IP addresses for a private workforce. Workers will only be able to
# login to their worker portal from an IP address within this range. By
# default, a workforce isn't restricted to specific IP addresses.
#
#
#
# [1]: https://docs.aws.amazon.com/vpc/latest/userguide/VPC_Subnets.html
#
# @option params [required, String] :workforce_name
# The name of the private workforce.
#
# @option params [Array] :tags
# An array of key-value pairs that contain metadata to help you
# categorize and organize our workforce. Each tag consists of a key and
# a value, both of which you define.
#
# @option params [Types::WorkforceVpcConfigRequest] :workforce_vpc_config
# Use this parameter to configure a workforce using VPC.
#
# @return [Types::CreateWorkforceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateWorkforceResponse#workforce_arn #workforce_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_workforce({
# cognito_config: {
# user_pool: "CognitoUserPool", # required
# client_id: "ClientId", # required
# },
# oidc_config: {
# client_id: "ClientId", # required
# client_secret: "ClientSecret", # required
# issuer: "OidcEndpoint", # required
# authorization_endpoint: "OidcEndpoint", # required
# token_endpoint: "OidcEndpoint", # required
# user_info_endpoint: "OidcEndpoint", # required
# logout_endpoint: "OidcEndpoint", # required
# jwks_uri: "OidcEndpoint", # required
# },
# source_ip_config: {
# cidrs: ["Cidr"], # required
# },
# workforce_name: "WorkforceName", # required
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# workforce_vpc_config: {
# vpc_id: "WorkforceVpcId",
# security_group_ids: ["WorkforceSecurityGroupId"],
# subnets: ["WorkforceSubnetId"],
# },
# })
#
# @example Response structure
#
# resp.workforce_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateWorkforce AWS API Documentation
#
# @overload create_workforce(params = {})
# @param [Hash] params ({})
def create_workforce(params = {}, options = {})
req = build_request(:create_workforce, params)
req.send_request(options)
end
# Creates a new work team for labeling your data. A work team is defined
# by one or more Amazon Cognito user pools. You must first create the
# user pools before you can create a work team.
#
# You cannot create more than 25 work teams in an account and region.
#
# @option params [required, String] :workteam_name
# The name of the work team. Use this name to identify the work team.
#
# @option params [String] :workforce_name
# The name of the workforce.
#
# @option params [required, Array] :member_definitions
# A list of `MemberDefinition` objects that contains objects that
# identify the workers that make up the work team.
#
# Workforces can be created using Amazon Cognito or your own OIDC
# Identity Provider (IdP). For private workforces created using Amazon
# Cognito use `CognitoMemberDefinition`. For workforces created using
# your own OIDC identity provider (IdP) use `OidcMemberDefinition`. Do
# not provide input for both of these parameters in a single request.
#
# For workforces created using Amazon Cognito, private work teams
# correspond to Amazon Cognito *user groups* within the user pool used
# to create a workforce. All of the `CognitoMemberDefinition` objects
# that make up the member definition must have the same `ClientId` and
# `UserPool` values. To add a Amazon Cognito user group to an existing
# worker pool, see [Adding groups to a User Pool](). For more
# information about user pools, see [Amazon Cognito User Pools][1].
#
# For workforces created using your own OIDC IdP, specify the user
# groups that you want to include in your private work team in
# `OidcMemberDefinition` by listing those groups in `Groups`.
#
#
#
# [1]: https://docs.aws.amazon.com/cognito/latest/developerguide/cognito-user-identity-pools.html
#
# @option params [required, String] :description
# A description of the work team.
#
# @option params [Types::NotificationConfiguration] :notification_configuration
# Configures notification of workers regarding available or expiring
# work items.
#
# @option params [Array] :tags
# An array of key-value pairs.
#
# For more information, see [Resource Tag][1] and [Using Cost Allocation
# Tags][2] in the Amazon Web Services Billing and Cost Management
# User Guide.
#
#
#
# [1]: https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/aws-properties-resource-tags.html
# [2]: https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/cost-alloc-tags.html#allocation-what
#
# @return [Types::CreateWorkteamResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::CreateWorkteamResponse#workteam_arn #workteam_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.create_workteam({
# workteam_name: "WorkteamName", # required
# workforce_name: "WorkforceName",
# member_definitions: [ # required
# {
# cognito_member_definition: {
# user_pool: "CognitoUserPool", # required
# user_group: "CognitoUserGroup", # required
# client_id: "ClientId", # required
# },
# oidc_member_definition: {
# groups: ["Group"], # required
# },
# },
# ],
# description: "String200", # required
# notification_configuration: {
# notification_topic_arn: "NotificationTopicArn",
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.workteam_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/CreateWorkteam AWS API Documentation
#
# @overload create_workteam(params = {})
# @param [Hash] params ({})
def create_workteam(params = {}, options = {})
req = build_request(:create_workteam, params)
req.send_request(options)
end
# Deletes an action.
#
# @option params [required, String] :action_name
# The name of the action to delete.
#
# @return [Types::DeleteActionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteActionResponse#action_arn #action_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_action({
# action_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.action_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteAction AWS API Documentation
#
# @overload delete_action(params = {})
# @param [Hash] params ({})
def delete_action(params = {}, options = {})
req = build_request(:delete_action, params)
req.send_request(options)
end
# Removes the specified algorithm from your account.
#
# @option params [required, String] :algorithm_name
# The name of the algorithm to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_algorithm({
# algorithm_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteAlgorithm AWS API Documentation
#
# @overload delete_algorithm(params = {})
# @param [Hash] params ({})
def delete_algorithm(params = {}, options = {})
req = build_request(:delete_algorithm, params)
req.send_request(options)
end
# Used to stop and delete an app.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [String] :user_profile_name
# The user profile name. If this value is not set, then `SpaceName` must
# be set.
#
# @option params [required, String] :app_type
# The type of app.
#
# @option params [required, String] :app_name
# The name of the app.
#
# @option params [String] :space_name
# The name of the space. If this value is not set, then
# `UserProfileName` must be set.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_app({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName",
# app_type: "JupyterServer", # required, accepts JupyterServer, KernelGateway, TensorBoard, RStudioServerPro, RSessionGateway
# app_name: "AppName", # required
# space_name: "SpaceName",
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteApp AWS API Documentation
#
# @overload delete_app(params = {})
# @param [Hash] params ({})
def delete_app(params = {}, options = {})
req = build_request(:delete_app, params)
req.send_request(options)
end
# Deletes an AppImageConfig.
#
# @option params [required, String] :app_image_config_name
# The name of the AppImageConfig to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_app_image_config({
# app_image_config_name: "AppImageConfigName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteAppImageConfig AWS API Documentation
#
# @overload delete_app_image_config(params = {})
# @param [Hash] params ({})
def delete_app_image_config(params = {}, options = {})
req = build_request(:delete_app_image_config, params)
req.send_request(options)
end
# Deletes an artifact. Either `ArtifactArn` or `Source` must be
# specified.
#
# @option params [String] :artifact_arn
# The Amazon Resource Name (ARN) of the artifact to delete.
#
# @option params [Types::ArtifactSource] :source
# The URI of the source.
#
# @return [Types::DeleteArtifactResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteArtifactResponse#artifact_arn #artifact_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_artifact({
# artifact_arn: "ArtifactArn",
# source: {
# source_uri: "String2048", # required
# source_types: [
# {
# source_id_type: "MD5Hash", # required, accepts MD5Hash, S3ETag, S3Version, Custom
# value: "String256", # required
# },
# ],
# },
# })
#
# @example Response structure
#
# resp.artifact_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteArtifact AWS API Documentation
#
# @overload delete_artifact(params = {})
# @param [Hash] params ({})
def delete_artifact(params = {}, options = {})
req = build_request(:delete_artifact, params)
req.send_request(options)
end
# Deletes an association.
#
# @option params [required, String] :source_arn
# The ARN of the source.
#
# @option params [required, String] :destination_arn
# The Amazon Resource Name (ARN) of the destination.
#
# @return [Types::DeleteAssociationResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteAssociationResponse#source_arn #source_arn} => String
# * {Types::DeleteAssociationResponse#destination_arn #destination_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_association({
# source_arn: "AssociationEntityArn", # required
# destination_arn: "AssociationEntityArn", # required
# })
#
# @example Response structure
#
# resp.source_arn #=> String
# resp.destination_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteAssociation AWS API Documentation
#
# @overload delete_association(params = {})
# @param [Hash] params ({})
def delete_association(params = {}, options = {})
req = build_request(:delete_association, params)
req.send_request(options)
end
# Deletes the specified Git repository from your account.
#
# @option params [required, String] :code_repository_name
# The name of the Git repository to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_code_repository({
# code_repository_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteCodeRepository AWS API Documentation
#
# @overload delete_code_repository(params = {})
# @param [Hash] params ({})
def delete_code_repository(params = {}, options = {})
req = build_request(:delete_code_repository, params)
req.send_request(options)
end
# Deletes an context.
#
# @option params [required, String] :context_name
# The name of the context to delete.
#
# @return [Types::DeleteContextResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteContextResponse#context_arn #context_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_context({
# context_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.context_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteContext AWS API Documentation
#
# @overload delete_context(params = {})
# @param [Hash] params ({})
def delete_context(params = {}, options = {})
req = build_request(:delete_context, params)
req.send_request(options)
end
# Deletes a data quality monitoring job definition.
#
# @option params [required, String] :job_definition_name
# The name of the data quality monitoring job definition to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_data_quality_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteDataQualityJobDefinition AWS API Documentation
#
# @overload delete_data_quality_job_definition(params = {})
# @param [Hash] params ({})
def delete_data_quality_job_definition(params = {}, options = {})
req = build_request(:delete_data_quality_job_definition, params)
req.send_request(options)
end
# Deletes a fleet.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_device_fleet({
# device_fleet_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteDeviceFleet AWS API Documentation
#
# @overload delete_device_fleet(params = {})
# @param [Hash] params ({})
def delete_device_fleet(params = {}, options = {})
req = build_request(:delete_device_fleet, params)
req.send_request(options)
end
# Used to delete a domain. If you onboarded with IAM mode, you will need
# to delete your domain to onboard again using IAM Identity Center. Use
# with caution. All of the members of the domain will lose access to
# their EFS volume, including data, notebooks, and other artifacts.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [Types::RetentionPolicy] :retention_policy
# The retention policy for this domain, which specifies whether
# resources will be retained after the Domain is deleted. By default,
# all resources are retained (not automatically deleted).
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_domain({
# domain_id: "DomainId", # required
# retention_policy: {
# home_efs_file_system: "Retain", # accepts Retain, Delete
# },
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteDomain AWS API Documentation
#
# @overload delete_domain(params = {})
# @param [Hash] params ({})
def delete_domain(params = {}, options = {})
req = build_request(:delete_domain, params)
req.send_request(options)
end
# Deletes an edge deployment plan if (and only if) all the stages in the
# plan are inactive or there are no stages in the plan.
#
# @option params [required, String] :edge_deployment_plan_name
# The name of the edge deployment plan to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_edge_deployment_plan({
# edge_deployment_plan_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteEdgeDeploymentPlan AWS API Documentation
#
# @overload delete_edge_deployment_plan(params = {})
# @param [Hash] params ({})
def delete_edge_deployment_plan(params = {}, options = {})
req = build_request(:delete_edge_deployment_plan, params)
req.send_request(options)
end
# Delete a stage in an edge deployment plan if (and only if) the stage
# is inactive.
#
# @option params [required, String] :edge_deployment_plan_name
# The name of the edge deployment plan from which the stage will be
# deleted.
#
# @option params [required, String] :stage_name
# The name of the stage.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_edge_deployment_stage({
# edge_deployment_plan_name: "EntityName", # required
# stage_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteEdgeDeploymentStage AWS API Documentation
#
# @overload delete_edge_deployment_stage(params = {})
# @param [Hash] params ({})
def delete_edge_deployment_stage(params = {}, options = {})
req = build_request(:delete_edge_deployment_stage, params)
req.send_request(options)
end
# Deletes an endpoint. SageMaker frees up all of the resources that were
# deployed when the endpoint was created.
#
# SageMaker retires any custom KMS key grants associated with the
# endpoint, meaning you don't need to use the [RevokeGrant][1] API
# call.
#
# When you delete your endpoint, SageMaker asynchronously deletes
# associated endpoint resources such as KMS key grants. You might still
# see these resources in your account for a few minutes after deleting
# your endpoint. Do not delete or revoke the permissions for your `
# ExecutionRoleArn `, otherwise SageMaker cannot delete these resources.
#
#
#
# [1]: http://docs.aws.amazon.com/kms/latest/APIReference/API_RevokeGrant.html
#
# @option params [required, String] :endpoint_name
# The name of the endpoint that you want to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_endpoint({
# endpoint_name: "EndpointName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteEndpoint AWS API Documentation
#
# @overload delete_endpoint(params = {})
# @param [Hash] params ({})
def delete_endpoint(params = {}, options = {})
req = build_request(:delete_endpoint, params)
req.send_request(options)
end
# Deletes an endpoint configuration. The `DeleteEndpointConfig` API
# deletes only the specified configuration. It does not delete endpoints
# created using the configuration.
#
# You must not delete an `EndpointConfig` in use by an endpoint that is
# live or while the `UpdateEndpoint` or `CreateEndpoint` operations are
# being performed on the endpoint. If you delete the `EndpointConfig` of
# an endpoint that is active or being created or updated you may lose
# visibility into the instance type the endpoint is using. The endpoint
# must be deleted in order to stop incurring charges.
#
# @option params [required, String] :endpoint_config_name
# The name of the endpoint configuration that you want to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_endpoint_config({
# endpoint_config_name: "EndpointConfigName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteEndpointConfig AWS API Documentation
#
# @overload delete_endpoint_config(params = {})
# @param [Hash] params ({})
def delete_endpoint_config(params = {}, options = {})
req = build_request(:delete_endpoint_config, params)
req.send_request(options)
end
# Deletes an SageMaker experiment. All trials associated with the
# experiment must be deleted first. Use the [ListTrials][1] API to get a
# list of the trials associated with the experiment.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ListTrials.html
#
# @option params [required, String] :experiment_name
# The name of the experiment to delete.
#
# @return [Types::DeleteExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteExperimentResponse#experiment_arn #experiment_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_experiment({
# experiment_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.experiment_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteExperiment AWS API Documentation
#
# @overload delete_experiment(params = {})
# @param [Hash] params ({})
def delete_experiment(params = {}, options = {})
req = build_request(:delete_experiment, params)
req.send_request(options)
end
# Delete the `FeatureGroup` and any data that was written to the
# `OnlineStore` of the `FeatureGroup`. Data cannot be accessed from the
# `OnlineStore` immediately after `DeleteFeatureGroup` is called.
#
# Data written into the `OfflineStore` will not be deleted. The Amazon
# Web Services Glue database and tables that are automatically created
# for your `OfflineStore` are not deleted.
#
# @option params [required, String] :feature_group_name
# The name of the `FeatureGroup` you want to delete. The name must be
# unique within an Amazon Web Services Region in an Amazon Web Services
# account.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_feature_group({
# feature_group_name: "FeatureGroupName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteFeatureGroup AWS API Documentation
#
# @overload delete_feature_group(params = {})
# @param [Hash] params ({})
def delete_feature_group(params = {}, options = {})
req = build_request(:delete_feature_group, params)
req.send_request(options)
end
# Deletes the specified flow definition.
#
# @option params [required, String] :flow_definition_name
# The name of the flow definition you are deleting.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_flow_definition({
# flow_definition_name: "FlowDefinitionName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteFlowDefinition AWS API Documentation
#
# @overload delete_flow_definition(params = {})
# @param [Hash] params ({})
def delete_flow_definition(params = {}, options = {})
req = build_request(:delete_flow_definition, params)
req.send_request(options)
end
# Delete a hub.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_hub({
# hub_name: "HubName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteHub AWS API Documentation
#
# @overload delete_hub(params = {})
# @param [Hash] params ({})
def delete_hub(params = {}, options = {})
req = build_request(:delete_hub, params)
req.send_request(options)
end
# Delete the contents of a hub.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub that you want to delete content in.
#
# @option params [required, String] :hub_content_type
# The type of content that you want to delete from a hub.
#
# @option params [required, String] :hub_content_name
# The name of the content that you want to delete from a hub.
#
# @option params [required, String] :hub_content_version
# The version of the content that you want to delete from a hub.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_hub_content({
# hub_name: "HubName", # required
# hub_content_type: "Model", # required, accepts Model, Notebook
# hub_content_name: "HubContentName", # required
# hub_content_version: "HubContentVersion", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteHubContent AWS API Documentation
#
# @overload delete_hub_content(params = {})
# @param [Hash] params ({})
def delete_hub_content(params = {}, options = {})
req = build_request(:delete_hub_content, params)
req.send_request(options)
end
# Use this operation to delete a human task user interface (worker task
# template).
#
# To see a list of human task user interfaces (work task templates) in
# your account, use [ListHumanTaskUis][1]. When you delete a worker task
# template, it no longer appears when you call `ListHumanTaskUis`.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ListHumanTaskUis.html
#
# @option params [required, String] :human_task_ui_name
# The name of the human task user interface (work task template) you
# want to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_human_task_ui({
# human_task_ui_name: "HumanTaskUiName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteHumanTaskUi AWS API Documentation
#
# @overload delete_human_task_ui(params = {})
# @param [Hash] params ({})
def delete_human_task_ui(params = {}, options = {})
req = build_request(:delete_human_task_ui, params)
req.send_request(options)
end
# Deletes a SageMaker image and all versions of the image. The container
# images aren't deleted.
#
# @option params [required, String] :image_name
# The name of the image to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_image({
# image_name: "ImageName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteImage AWS API Documentation
#
# @overload delete_image(params = {})
# @param [Hash] params ({})
def delete_image(params = {}, options = {})
req = build_request(:delete_image, params)
req.send_request(options)
end
# Deletes a version of a SageMaker image. The container image the
# version represents isn't deleted.
#
# @option params [required, String] :image_name
# The name of the image to delete.
#
# @option params [Integer] :version
# The version to delete.
#
# @option params [String] :alias
# The alias of the image to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_image_version({
# image_name: "ImageName", # required
# version: 1,
# alias: "SageMakerImageVersionAlias",
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteImageVersion AWS API Documentation
#
# @overload delete_image_version(params = {})
# @param [Hash] params ({})
def delete_image_version(params = {}, options = {})
req = build_request(:delete_image_version, params)
req.send_request(options)
end
# Deletes an inference experiment.
#
# This operation does not delete your endpoint, variants, or any
# underlying resources. This operation only deletes the metadata of your
# experiment.
#
#
#
# @option params [required, String] :name
# The name of the inference experiment you want to delete.
#
# @return [Types::DeleteInferenceExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteInferenceExperimentResponse#inference_experiment_arn #inference_experiment_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_inference_experiment({
# name: "InferenceExperimentName", # required
# })
#
# @example Response structure
#
# resp.inference_experiment_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteInferenceExperiment AWS API Documentation
#
# @overload delete_inference_experiment(params = {})
# @param [Hash] params ({})
def delete_inference_experiment(params = {}, options = {})
req = build_request(:delete_inference_experiment, params)
req.send_request(options)
end
# Deletes a model. The `DeleteModel` API deletes only the model entry
# that was created in SageMaker when you called the `CreateModel` API.
# It does not delete model artifacts, inference code, or the IAM role
# that you specified when creating the model.
#
# @option params [required, String] :model_name
# The name of the model to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model({
# model_name: "ModelName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModel AWS API Documentation
#
# @overload delete_model(params = {})
# @param [Hash] params ({})
def delete_model(params = {}, options = {})
req = build_request(:delete_model, params)
req.send_request(options)
end
# Deletes an Amazon SageMaker model bias job definition.
#
# @option params [required, String] :job_definition_name
# The name of the model bias job definition to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_bias_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelBiasJobDefinition AWS API Documentation
#
# @overload delete_model_bias_job_definition(params = {})
# @param [Hash] params ({})
def delete_model_bias_job_definition(params = {}, options = {})
req = build_request(:delete_model_bias_job_definition, params)
req.send_request(options)
end
# Deletes an Amazon SageMaker Model Card.
#
# @option params [required, String] :model_card_name
# The name of the model card to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_card({
# model_card_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelCard AWS API Documentation
#
# @overload delete_model_card(params = {})
# @param [Hash] params ({})
def delete_model_card(params = {}, options = {})
req = build_request(:delete_model_card, params)
req.send_request(options)
end
# Deletes an Amazon SageMaker model explainability job definition.
#
# @option params [required, String] :job_definition_name
# The name of the model explainability job definition to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_explainability_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelExplainabilityJobDefinition AWS API Documentation
#
# @overload delete_model_explainability_job_definition(params = {})
# @param [Hash] params ({})
def delete_model_explainability_job_definition(params = {}, options = {})
req = build_request(:delete_model_explainability_job_definition, params)
req.send_request(options)
end
# Deletes a model package.
#
# A model package is used to create SageMaker models or list on Amazon
# Web Services Marketplace. Buyers can subscribe to model packages
# listed on Amazon Web Services Marketplace to create models in
# SageMaker.
#
# @option params [required, String] :model_package_name
# The name or Amazon Resource Name (ARN) of the model package to delete.
#
# When you specify a name, the name must have 1 to 63 characters. Valid
# characters are a-z, A-Z, 0-9, and - (hyphen).
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_package({
# model_package_name: "VersionedArnOrName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelPackage AWS API Documentation
#
# @overload delete_model_package(params = {})
# @param [Hash] params ({})
def delete_model_package(params = {}, options = {})
req = build_request(:delete_model_package, params)
req.send_request(options)
end
# Deletes the specified model group.
#
# @option params [required, String] :model_package_group_name
# The name of the model group to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_package_group({
# model_package_group_name: "ArnOrName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelPackageGroup AWS API Documentation
#
# @overload delete_model_package_group(params = {})
# @param [Hash] params ({})
def delete_model_package_group(params = {}, options = {})
req = build_request(:delete_model_package_group, params)
req.send_request(options)
end
# Deletes a model group resource policy.
#
# @option params [required, String] :model_package_group_name
# The name of the model group for which to delete the policy.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_package_group_policy({
# model_package_group_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelPackageGroupPolicy AWS API Documentation
#
# @overload delete_model_package_group_policy(params = {})
# @param [Hash] params ({})
def delete_model_package_group_policy(params = {}, options = {})
req = build_request(:delete_model_package_group_policy, params)
req.send_request(options)
end
# Deletes the secified model quality monitoring job definition.
#
# @option params [required, String] :job_definition_name
# The name of the model quality monitoring job definition to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_model_quality_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteModelQualityJobDefinition AWS API Documentation
#
# @overload delete_model_quality_job_definition(params = {})
# @param [Hash] params ({})
def delete_model_quality_job_definition(params = {}, options = {})
req = build_request(:delete_model_quality_job_definition, params)
req.send_request(options)
end
# Deletes a monitoring schedule. Also stops the schedule had not already
# been stopped. This does not delete the job execution history of the
# monitoring schedule.
#
# @option params [required, String] :monitoring_schedule_name
# The name of the monitoring schedule to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_monitoring_schedule({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteMonitoringSchedule AWS API Documentation
#
# @overload delete_monitoring_schedule(params = {})
# @param [Hash] params ({})
def delete_monitoring_schedule(params = {}, options = {})
req = build_request(:delete_monitoring_schedule, params)
req.send_request(options)
end
# Deletes an SageMaker notebook instance. Before you can delete a
# notebook instance, you must call the `StopNotebookInstance` API.
#
# When you delete a notebook instance, you lose all of your data.
# SageMaker removes the ML compute instance, and deletes the ML storage
# volume and the network interface associated with the notebook
# instance.
#
# @option params [required, String] :notebook_instance_name
# The name of the SageMaker notebook instance to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_notebook_instance({
# notebook_instance_name: "NotebookInstanceName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteNotebookInstance AWS API Documentation
#
# @overload delete_notebook_instance(params = {})
# @param [Hash] params ({})
def delete_notebook_instance(params = {}, options = {})
req = build_request(:delete_notebook_instance, params)
req.send_request(options)
end
# Deletes a notebook instance lifecycle configuration.
#
# @option params [required, String] :notebook_instance_lifecycle_config_name
# The name of the lifecycle configuration to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_notebook_instance_lifecycle_config({
# notebook_instance_lifecycle_config_name: "NotebookInstanceLifecycleConfigName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteNotebookInstanceLifecycleConfig AWS API Documentation
#
# @overload delete_notebook_instance_lifecycle_config(params = {})
# @param [Hash] params ({})
def delete_notebook_instance_lifecycle_config(params = {}, options = {})
req = build_request(:delete_notebook_instance_lifecycle_config, params)
req.send_request(options)
end
# Deletes a pipeline if there are no running instances of the pipeline.
# To delete a pipeline, you must stop all running instances of the
# pipeline using the `StopPipelineExecution` API. When you delete a
# pipeline, all instances of the pipeline are deleted.
#
# @option params [required, String] :pipeline_name
# The name of the pipeline to delete.
#
# @option params [required, String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than one time.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @return [Types::DeletePipelineResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeletePipelineResponse#pipeline_arn #pipeline_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_pipeline({
# pipeline_name: "PipelineName", # required
# client_request_token: "IdempotencyToken", # required
# })
#
# @example Response structure
#
# resp.pipeline_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeletePipeline AWS API Documentation
#
# @overload delete_pipeline(params = {})
# @param [Hash] params ({})
def delete_pipeline(params = {}, options = {})
req = build_request(:delete_pipeline, params)
req.send_request(options)
end
# Delete the specified project.
#
# @option params [required, String] :project_name
# The name of the project to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_project({
# project_name: "ProjectEntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteProject AWS API Documentation
#
# @overload delete_project(params = {})
# @param [Hash] params ({})
def delete_project(params = {}, options = {})
req = build_request(:delete_project, params)
req.send_request(options)
end
# Used to delete a space.
#
# @option params [required, String] :domain_id
# The ID of the associated Domain.
#
# @option params [required, String] :space_name
# The name of the space.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_space({
# domain_id: "DomainId", # required
# space_name: "SpaceName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteSpace AWS API Documentation
#
# @overload delete_space(params = {})
# @param [Hash] params ({})
def delete_space(params = {}, options = {})
req = build_request(:delete_space, params)
req.send_request(options)
end
# Deletes the Studio Lifecycle Configuration. In order to delete the
# Lifecycle Configuration, there must be no running apps using the
# Lifecycle Configuration. You must also remove the Lifecycle
# Configuration from UserSettings in all Domains and UserProfiles.
#
# @option params [required, String] :studio_lifecycle_config_name
# The name of the Studio Lifecycle Configuration to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_studio_lifecycle_config({
# studio_lifecycle_config_name: "StudioLifecycleConfigName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteStudioLifecycleConfig AWS API Documentation
#
# @overload delete_studio_lifecycle_config(params = {})
# @param [Hash] params ({})
def delete_studio_lifecycle_config(params = {}, options = {})
req = build_request(:delete_studio_lifecycle_config, params)
req.send_request(options)
end
# Deletes the specified tags from an SageMaker resource.
#
# To list a resource's tags, use the `ListTags` API.
#
# When you call this API to delete tags from a hyperparameter tuning
# job, the deleted tags are not removed from training jobs that the
# hyperparameter tuning job launched before you called this API.
#
#
#
# When you call this API to delete tags from a SageMaker Studio Domain
# or User Profile, the deleted tags are not removed from Apps that the
# SageMaker Studio Domain or User Profile launched before you called
# this API.
#
#
#
# @option params [required, String] :resource_arn
# The Amazon Resource Name (ARN) of the resource whose tags you want to
# delete.
#
# @option params [required, Array] :tag_keys
# An array or one or more tag keys to delete.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_tags({
# resource_arn: "ResourceArn", # required
# tag_keys: ["TagKey"], # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteTags AWS API Documentation
#
# @overload delete_tags(params = {})
# @param [Hash] params ({})
def delete_tags(params = {}, options = {})
req = build_request(:delete_tags, params)
req.send_request(options)
end
# Deletes the specified trial. All trial components that make up the
# trial must be deleted first. Use the [DescribeTrialComponent][1] API
# to get the list of trial components.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeTrialComponent.html
#
# @option params [required, String] :trial_name
# The name of the trial to delete.
#
# @return [Types::DeleteTrialResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteTrialResponse#trial_arn #trial_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_trial({
# trial_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.trial_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteTrial AWS API Documentation
#
# @overload delete_trial(params = {})
# @param [Hash] params ({})
def delete_trial(params = {}, options = {})
req = build_request(:delete_trial, params)
req.send_request(options)
end
# Deletes the specified trial component. A trial component must be
# disassociated from all trials before the trial component can be
# deleted. To disassociate a trial component from a trial, call the
# [DisassociateTrialComponent][1] API.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DisassociateTrialComponent.html
#
# @option params [required, String] :trial_component_name
# The name of the component to delete.
#
# @return [Types::DeleteTrialComponentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteTrialComponentResponse#trial_component_arn #trial_component_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.delete_trial_component({
# trial_component_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.trial_component_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteTrialComponent AWS API Documentation
#
# @overload delete_trial_component(params = {})
# @param [Hash] params ({})
def delete_trial_component(params = {}, options = {})
req = build_request(:delete_trial_component, params)
req.send_request(options)
end
# Deletes a user profile. When a user profile is deleted, the user loses
# access to their EFS volume, including data, notebooks, and other
# artifacts.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [required, String] :user_profile_name
# The user profile name.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_user_profile({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteUserProfile AWS API Documentation
#
# @overload delete_user_profile(params = {})
# @param [Hash] params ({})
def delete_user_profile(params = {}, options = {})
req = build_request(:delete_user_profile, params)
req.send_request(options)
end
# Use this operation to delete a workforce.
#
# If you want to create a new workforce in an Amazon Web Services Region
# where a workforce already exists, use this operation to delete the
# existing workforce and then use [CreateWorkforce][1] to create a new
# workforce.
#
# If a private workforce contains one or more work teams, you must use
# the [DeleteWorkteam][2] operation to delete all work teams before you
# delete the workforce. If you try to delete a workforce that contains
# one or more work teams, you will recieve a `ResourceInUse` error.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateWorkforce.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DeleteWorkteam.html
#
# @option params [required, String] :workforce_name
# The name of the workforce.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.delete_workforce({
# workforce_name: "WorkforceName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteWorkforce AWS API Documentation
#
# @overload delete_workforce(params = {})
# @param [Hash] params ({})
def delete_workforce(params = {}, options = {})
req = build_request(:delete_workforce, params)
req.send_request(options)
end
# Deletes an existing work team. This operation can't be undone.
#
# @option params [required, String] :workteam_name
# The name of the work team to delete.
#
# @return [Types::DeleteWorkteamResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DeleteWorkteamResponse#success #success} => Boolean
#
# @example Request syntax with placeholder values
#
# resp = client.delete_workteam({
# workteam_name: "WorkteamName", # required
# })
#
# @example Response structure
#
# resp.success #=> Boolean
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeleteWorkteam AWS API Documentation
#
# @overload delete_workteam(params = {})
# @param [Hash] params ({})
def delete_workteam(params = {}, options = {})
req = build_request(:delete_workteam, params)
req.send_request(options)
end
# Deregisters the specified devices. After you deregister a device, you
# will need to re-register the devices.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet the devices belong to.
#
# @option params [required, Array] :device_names
# The unique IDs of the devices.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.deregister_devices({
# device_fleet_name: "EntityName", # required
# device_names: ["DeviceName"], # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DeregisterDevices AWS API Documentation
#
# @overload deregister_devices(params = {})
# @param [Hash] params ({})
def deregister_devices(params = {}, options = {})
req = build_request(:deregister_devices, params)
req.send_request(options)
end
# Describes an action.
#
# @option params [required, String] :action_name
# The name of the action to describe.
#
# @return [Types::DescribeActionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeActionResponse#action_name #action_name} => String
# * {Types::DescribeActionResponse#action_arn #action_arn} => String
# * {Types::DescribeActionResponse#source #source} => Types::ActionSource
# * {Types::DescribeActionResponse#action_type #action_type} => String
# * {Types::DescribeActionResponse#description #description} => String
# * {Types::DescribeActionResponse#status #status} => String
# * {Types::DescribeActionResponse#properties #properties} => Hash<String,String>
# * {Types::DescribeActionResponse#creation_time #creation_time} => Time
# * {Types::DescribeActionResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeActionResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeActionResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeActionResponse#metadata_properties #metadata_properties} => Types::MetadataProperties
# * {Types::DescribeActionResponse#lineage_group_arn #lineage_group_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_action({
# action_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.action_name #=> String
# resp.action_arn #=> String
# resp.source.source_uri #=> String
# resp.source.source_type #=> String
# resp.source.source_id #=> String
# resp.action_type #=> String
# resp.description #=> String
# resp.status #=> String, one of "Unknown", "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.properties #=> Hash
# resp.properties["StringParameterValue"] #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.metadata_properties.commit_id #=> String
# resp.metadata_properties.repository #=> String
# resp.metadata_properties.generated_by #=> String
# resp.metadata_properties.project_id #=> String
# resp.lineage_group_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeAction AWS API Documentation
#
# @overload describe_action(params = {})
# @param [Hash] params ({})
def describe_action(params = {}, options = {})
req = build_request(:describe_action, params)
req.send_request(options)
end
# Returns a description of the specified algorithm that is in your
# account.
#
# @option params [required, String] :algorithm_name
# The name of the algorithm to describe.
#
# @return [Types::DescribeAlgorithmOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeAlgorithmOutput#algorithm_name #algorithm_name} => String
# * {Types::DescribeAlgorithmOutput#algorithm_arn #algorithm_arn} => String
# * {Types::DescribeAlgorithmOutput#algorithm_description #algorithm_description} => String
# * {Types::DescribeAlgorithmOutput#creation_time #creation_time} => Time
# * {Types::DescribeAlgorithmOutput#training_specification #training_specification} => Types::TrainingSpecification
# * {Types::DescribeAlgorithmOutput#inference_specification #inference_specification} => Types::InferenceSpecification
# * {Types::DescribeAlgorithmOutput#validation_specification #validation_specification} => Types::AlgorithmValidationSpecification
# * {Types::DescribeAlgorithmOutput#algorithm_status #algorithm_status} => String
# * {Types::DescribeAlgorithmOutput#algorithm_status_details #algorithm_status_details} => Types::AlgorithmStatusDetails
# * {Types::DescribeAlgorithmOutput#product_id #product_id} => String
# * {Types::DescribeAlgorithmOutput#certify_for_marketplace #certify_for_marketplace} => Boolean
#
# @example Request syntax with placeholder values
#
# resp = client.describe_algorithm({
# algorithm_name: "ArnOrName", # required
# })
#
# @example Response structure
#
# resp.algorithm_name #=> String
# resp.algorithm_arn #=> String
# resp.algorithm_description #=> String
# resp.creation_time #=> Time
# resp.training_specification.training_image #=> String
# resp.training_specification.training_image_digest #=> String
# resp.training_specification.supported_hyper_parameters #=> Array
# resp.training_specification.supported_hyper_parameters[0].name #=> String
# resp.training_specification.supported_hyper_parameters[0].description #=> String
# resp.training_specification.supported_hyper_parameters[0].type #=> String, one of "Integer", "Continuous", "Categorical", "FreeText"
# resp.training_specification.supported_hyper_parameters[0].range.integer_parameter_range_specification.min_value #=> String
# resp.training_specification.supported_hyper_parameters[0].range.integer_parameter_range_specification.max_value #=> String
# resp.training_specification.supported_hyper_parameters[0].range.continuous_parameter_range_specification.min_value #=> String
# resp.training_specification.supported_hyper_parameters[0].range.continuous_parameter_range_specification.max_value #=> String
# resp.training_specification.supported_hyper_parameters[0].range.categorical_parameter_range_specification.values #=> Array
# resp.training_specification.supported_hyper_parameters[0].range.categorical_parameter_range_specification.values[0] #=> String
# resp.training_specification.supported_hyper_parameters[0].is_tunable #=> Boolean
# resp.training_specification.supported_hyper_parameters[0].is_required #=> Boolean
# resp.training_specification.supported_hyper_parameters[0].default_value #=> String
# resp.training_specification.supported_training_instance_types #=> Array
# resp.training_specification.supported_training_instance_types[0] #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_specification.supports_distributed_training #=> Boolean
# resp.training_specification.metric_definitions #=> Array
# resp.training_specification.metric_definitions[0].name #=> String
# resp.training_specification.metric_definitions[0].regex #=> String
# resp.training_specification.training_channels #=> Array
# resp.training_specification.training_channels[0].name #=> String
# resp.training_specification.training_channels[0].description #=> String
# resp.training_specification.training_channels[0].is_required #=> Boolean
# resp.training_specification.training_channels[0].supported_content_types #=> Array
# resp.training_specification.training_channels[0].supported_content_types[0] #=> String
# resp.training_specification.training_channels[0].supported_compression_types #=> Array
# resp.training_specification.training_channels[0].supported_compression_types[0] #=> String, one of "None", "Gzip"
# resp.training_specification.training_channels[0].supported_input_modes #=> Array
# resp.training_specification.training_channels[0].supported_input_modes[0] #=> String, one of "Pipe", "File", "FastFile"
# resp.training_specification.supported_tuning_job_objective_metrics #=> Array
# resp.training_specification.supported_tuning_job_objective_metrics[0].type #=> String, one of "Maximize", "Minimize"
# resp.training_specification.supported_tuning_job_objective_metrics[0].metric_name #=> String
# resp.inference_specification.containers #=> Array
# resp.inference_specification.containers[0].container_hostname #=> String
# resp.inference_specification.containers[0].image #=> String
# resp.inference_specification.containers[0].image_digest #=> String
# resp.inference_specification.containers[0].model_data_url #=> String
# resp.inference_specification.containers[0].product_id #=> String
# resp.inference_specification.containers[0].environment #=> Hash
# resp.inference_specification.containers[0].environment["EnvironmentKey"] #=> String
# resp.inference_specification.containers[0].model_input.data_input_config #=> String
# resp.inference_specification.containers[0].framework #=> String
# resp.inference_specification.containers[0].framework_version #=> String
# resp.inference_specification.containers[0].nearest_model_name #=> String
# resp.inference_specification.supported_transform_instance_types #=> Array
# resp.inference_specification.supported_transform_instance_types[0] #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.inference_specification.supported_realtime_inference_instance_types #=> Array
# resp.inference_specification.supported_realtime_inference_instance_types[0] #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.inference_specification.supported_content_types #=> Array
# resp.inference_specification.supported_content_types[0] #=> String
# resp.inference_specification.supported_response_mime_types #=> Array
# resp.inference_specification.supported_response_mime_types[0] #=> String
# resp.validation_specification.validation_role #=> String
# resp.validation_specification.validation_profiles #=> Array
# resp.validation_specification.validation_profiles[0].profile_name #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.training_input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.validation_specification.validation_profiles[0].training_job_definition.hyper_parameters #=> Hash
# resp.validation_specification.validation_profiles[0].training_job_definition.hyper_parameters["HyperParameterKey"] #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config #=> Array
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].channel_name #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.s3_uri #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.attribute_names #=> Array
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.attribute_names[0] #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.instance_group_names #=> Array
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.s3_data_source.instance_group_names[0] #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.file_system_data_source.file_system_id #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.file_system_data_source.file_system_access_mode #=> String, one of "rw", "ro"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.file_system_data_source.file_system_type #=> String, one of "EFS", "FSxLustre"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].data_source.file_system_data_source.directory_path #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].content_type #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].compression_type #=> String, one of "None", "Gzip"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].record_wrapper_type #=> String, one of "None", "RecordIO"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.validation_specification.validation_profiles[0].training_job_definition.input_data_config[0].shuffle_config.seed #=> Integer
# resp.validation_specification.validation_profiles[0].training_job_definition.output_data_config.kms_key_id #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.output_data_config.s3_output_path #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.instance_count #=> Integer
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.volume_size_in_gb #=> Integer
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.volume_kms_key_id #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.instance_groups #=> Array
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.instance_groups[0].instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.instance_groups[0].instance_count #=> Integer
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.instance_groups[0].instance_group_name #=> String
# resp.validation_specification.validation_profiles[0].training_job_definition.resource_config.keep_alive_period_in_seconds #=> Integer
# resp.validation_specification.validation_profiles[0].training_job_definition.stopping_condition.max_runtime_in_seconds #=> Integer
# resp.validation_specification.validation_profiles[0].training_job_definition.stopping_condition.max_wait_time_in_seconds #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.max_concurrent_transforms #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.max_payload_in_mb #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.batch_strategy #=> String, one of "MultiRecord", "SingleRecord"
# resp.validation_specification.validation_profiles[0].transform_job_definition.environment #=> Hash
# resp.validation_specification.validation_profiles[0].transform_job_definition.environment["TransformEnvironmentKey"] #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.data_source.s3_data_source.s3_uri #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.content_type #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.compression_type #=> String, one of "None", "Gzip"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.split_type #=> String, one of "None", "Line", "RecordIO", "TFRecord"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.s3_output_path #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.accept #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.assemble_with #=> String, one of "None", "Line"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.kms_key_id #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_resources.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_resources.instance_count #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_resources.volume_kms_key_id #=> String
# resp.algorithm_status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Deleting"
# resp.algorithm_status_details.validation_statuses #=> Array
# resp.algorithm_status_details.validation_statuses[0].name #=> String
# resp.algorithm_status_details.validation_statuses[0].status #=> String, one of "NotStarted", "InProgress", "Completed", "Failed"
# resp.algorithm_status_details.validation_statuses[0].failure_reason #=> String
# resp.algorithm_status_details.image_scan_statuses #=> Array
# resp.algorithm_status_details.image_scan_statuses[0].name #=> String
# resp.algorithm_status_details.image_scan_statuses[0].status #=> String, one of "NotStarted", "InProgress", "Completed", "Failed"
# resp.algorithm_status_details.image_scan_statuses[0].failure_reason #=> String
# resp.product_id #=> String
# resp.certify_for_marketplace #=> Boolean
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeAlgorithm AWS API Documentation
#
# @overload describe_algorithm(params = {})
# @param [Hash] params ({})
def describe_algorithm(params = {}, options = {})
req = build_request(:describe_algorithm, params)
req.send_request(options)
end
# Describes the app.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [String] :user_profile_name
# The user profile name. If this value is not set, then `SpaceName` must
# be set.
#
# @option params [required, String] :app_type
# The type of app.
#
# @option params [required, String] :app_name
# The name of the app.
#
# @option params [String] :space_name
# The name of the space.
#
# @return [Types::DescribeAppResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeAppResponse#app_arn #app_arn} => String
# * {Types::DescribeAppResponse#app_type #app_type} => String
# * {Types::DescribeAppResponse#app_name #app_name} => String
# * {Types::DescribeAppResponse#domain_id #domain_id} => String
# * {Types::DescribeAppResponse#user_profile_name #user_profile_name} => String
# * {Types::DescribeAppResponse#status #status} => String
# * {Types::DescribeAppResponse#last_health_check_timestamp #last_health_check_timestamp} => Time
# * {Types::DescribeAppResponse#last_user_activity_timestamp #last_user_activity_timestamp} => Time
# * {Types::DescribeAppResponse#creation_time #creation_time} => Time
# * {Types::DescribeAppResponse#failure_reason #failure_reason} => String
# * {Types::DescribeAppResponse#resource_spec #resource_spec} => Types::ResourceSpec
# * {Types::DescribeAppResponse#space_name #space_name} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_app({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName",
# app_type: "JupyterServer", # required, accepts JupyterServer, KernelGateway, TensorBoard, RStudioServerPro, RSessionGateway
# app_name: "AppName", # required
# space_name: "SpaceName",
# })
#
# @example Response structure
#
# resp.app_arn #=> String
# resp.app_type #=> String, one of "JupyterServer", "KernelGateway", "TensorBoard", "RStudioServerPro", "RSessionGateway"
# resp.app_name #=> String
# resp.domain_id #=> String
# resp.user_profile_name #=> String
# resp.status #=> String, one of "Deleted", "Deleting", "Failed", "InService", "Pending"
# resp.last_health_check_timestamp #=> Time
# resp.last_user_activity_timestamp #=> Time
# resp.creation_time #=> Time
# resp.failure_reason #=> String
# resp.resource_spec.sage_maker_image_arn #=> String
# resp.resource_spec.sage_maker_image_version_arn #=> String
# resp.resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.resource_spec.lifecycle_config_arn #=> String
# resp.space_name #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeApp AWS API Documentation
#
# @overload describe_app(params = {})
# @param [Hash] params ({})
def describe_app(params = {}, options = {})
req = build_request(:describe_app, params)
req.send_request(options)
end
# Describes an AppImageConfig.
#
# @option params [required, String] :app_image_config_name
# The name of the AppImageConfig to describe.
#
# @return [Types::DescribeAppImageConfigResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeAppImageConfigResponse#app_image_config_arn #app_image_config_arn} => String
# * {Types::DescribeAppImageConfigResponse#app_image_config_name #app_image_config_name} => String
# * {Types::DescribeAppImageConfigResponse#creation_time #creation_time} => Time
# * {Types::DescribeAppImageConfigResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeAppImageConfigResponse#kernel_gateway_image_config #kernel_gateway_image_config} => Types::KernelGatewayImageConfig
#
# @example Request syntax with placeholder values
#
# resp = client.describe_app_image_config({
# app_image_config_name: "AppImageConfigName", # required
# })
#
# @example Response structure
#
# resp.app_image_config_arn #=> String
# resp.app_image_config_name #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.kernel_gateway_image_config.kernel_specs #=> Array
# resp.kernel_gateway_image_config.kernel_specs[0].name #=> String
# resp.kernel_gateway_image_config.kernel_specs[0].display_name #=> String
# resp.kernel_gateway_image_config.file_system_config.mount_path #=> String
# resp.kernel_gateway_image_config.file_system_config.default_uid #=> Integer
# resp.kernel_gateway_image_config.file_system_config.default_gid #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeAppImageConfig AWS API Documentation
#
# @overload describe_app_image_config(params = {})
# @param [Hash] params ({})
def describe_app_image_config(params = {}, options = {})
req = build_request(:describe_app_image_config, params)
req.send_request(options)
end
# Describes an artifact.
#
# @option params [required, String] :artifact_arn
# The Amazon Resource Name (ARN) of the artifact to describe.
#
# @return [Types::DescribeArtifactResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeArtifactResponse#artifact_name #artifact_name} => String
# * {Types::DescribeArtifactResponse#artifact_arn #artifact_arn} => String
# * {Types::DescribeArtifactResponse#source #source} => Types::ArtifactSource
# * {Types::DescribeArtifactResponse#artifact_type #artifact_type} => String
# * {Types::DescribeArtifactResponse#properties #properties} => Hash<String,String>
# * {Types::DescribeArtifactResponse#creation_time #creation_time} => Time
# * {Types::DescribeArtifactResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeArtifactResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeArtifactResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeArtifactResponse#metadata_properties #metadata_properties} => Types::MetadataProperties
# * {Types::DescribeArtifactResponse#lineage_group_arn #lineage_group_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_artifact({
# artifact_arn: "ArtifactArn", # required
# })
#
# @example Response structure
#
# resp.artifact_name #=> String
# resp.artifact_arn #=> String
# resp.source.source_uri #=> String
# resp.source.source_types #=> Array
# resp.source.source_types[0].source_id_type #=> String, one of "MD5Hash", "S3ETag", "S3Version", "Custom"
# resp.source.source_types[0].value #=> String
# resp.artifact_type #=> String
# resp.properties #=> Hash
# resp.properties["StringParameterValue"] #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.metadata_properties.commit_id #=> String
# resp.metadata_properties.repository #=> String
# resp.metadata_properties.generated_by #=> String
# resp.metadata_properties.project_id #=> String
# resp.lineage_group_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeArtifact AWS API Documentation
#
# @overload describe_artifact(params = {})
# @param [Hash] params ({})
def describe_artifact(params = {}, options = {})
req = build_request(:describe_artifact, params)
req.send_request(options)
end
# Returns information about an Amazon SageMaker AutoML job.
#
# @option params [required, String] :auto_ml_job_name
# Requests information about an AutoML job using its unique name.
#
# @return [Types::DescribeAutoMLJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_name #auto_ml_job_name} => String
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_arn #auto_ml_job_arn} => String
# * {Types::DescribeAutoMLJobResponse#input_data_config #input_data_config} => Array<Types::AutoMLChannel>
# * {Types::DescribeAutoMLJobResponse#output_data_config #output_data_config} => Types::AutoMLOutputDataConfig
# * {Types::DescribeAutoMLJobResponse#role_arn #role_arn} => String
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_objective #auto_ml_job_objective} => Types::AutoMLJobObjective
# * {Types::DescribeAutoMLJobResponse#problem_type #problem_type} => String
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_config #auto_ml_job_config} => Types::AutoMLJobConfig
# * {Types::DescribeAutoMLJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeAutoMLJobResponse#end_time #end_time} => Time
# * {Types::DescribeAutoMLJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeAutoMLJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeAutoMLJobResponse#partial_failure_reasons #partial_failure_reasons} => Array<Types::AutoMLPartialFailureReason>
# * {Types::DescribeAutoMLJobResponse#best_candidate #best_candidate} => Types::AutoMLCandidate
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_status #auto_ml_job_status} => String
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_secondary_status #auto_ml_job_secondary_status} => String
# * {Types::DescribeAutoMLJobResponse#generate_candidate_definitions_only #generate_candidate_definitions_only} => Boolean
# * {Types::DescribeAutoMLJobResponse#auto_ml_job_artifacts #auto_ml_job_artifacts} => Types::AutoMLJobArtifacts
# * {Types::DescribeAutoMLJobResponse#resolved_attributes #resolved_attributes} => Types::ResolvedAttributes
# * {Types::DescribeAutoMLJobResponse#model_deploy_config #model_deploy_config} => Types::ModelDeployConfig
# * {Types::DescribeAutoMLJobResponse#model_deploy_result #model_deploy_result} => Types::ModelDeployResult
#
# @example Request syntax with placeholder values
#
# resp = client.describe_auto_ml_job({
# auto_ml_job_name: "AutoMLJobName", # required
# })
#
# @example Response structure
#
# resp.auto_ml_job_name #=> String
# resp.auto_ml_job_arn #=> String
# resp.input_data_config #=> Array
# resp.input_data_config[0].data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.input_data_config[0].data_source.s3_data_source.s3_uri #=> String
# resp.input_data_config[0].compression_type #=> String, one of "None", "Gzip"
# resp.input_data_config[0].target_attribute_name #=> String
# resp.input_data_config[0].content_type #=> String
# resp.input_data_config[0].channel_type #=> String, one of "training", "validation"
# resp.input_data_config[0].sample_weight_attribute_name #=> String
# resp.output_data_config.kms_key_id #=> String
# resp.output_data_config.s3_output_path #=> String
# resp.role_arn #=> String
# resp.auto_ml_job_objective.metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.problem_type #=> String, one of "BinaryClassification", "MulticlassClassification", "Regression"
# resp.auto_ml_job_config.completion_criteria.max_candidates #=> Integer
# resp.auto_ml_job_config.completion_criteria.max_runtime_per_training_job_in_seconds #=> Integer
# resp.auto_ml_job_config.completion_criteria.max_auto_ml_job_runtime_in_seconds #=> Integer
# resp.auto_ml_job_config.security_config.volume_kms_key_id #=> String
# resp.auto_ml_job_config.security_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.auto_ml_job_config.security_config.vpc_config.security_group_ids #=> Array
# resp.auto_ml_job_config.security_config.vpc_config.security_group_ids[0] #=> String
# resp.auto_ml_job_config.security_config.vpc_config.subnets #=> Array
# resp.auto_ml_job_config.security_config.vpc_config.subnets[0] #=> String
# resp.auto_ml_job_config.data_split_config.validation_fraction #=> Float
# resp.auto_ml_job_config.candidate_generation_config.feature_specification_s3_uri #=> String
# resp.auto_ml_job_config.candidate_generation_config.algorithms_config #=> Array
# resp.auto_ml_job_config.candidate_generation_config.algorithms_config[0].auto_ml_algorithms #=> Array
# resp.auto_ml_job_config.candidate_generation_config.algorithms_config[0].auto_ml_algorithms[0] #=> String, one of "xgboost", "linear-learner", "mlp", "lightgbm", "catboost", "randomforest", "extra-trees", "nn-torch", "fastai"
# resp.auto_ml_job_config.mode #=> String, one of "AUTO", "ENSEMBLING", "HYPERPARAMETER_TUNING"
# resp.creation_time #=> Time
# resp.end_time #=> Time
# resp.last_modified_time #=> Time
# resp.failure_reason #=> String
# resp.partial_failure_reasons #=> Array
# resp.partial_failure_reasons[0].partial_failure_message #=> String
# resp.best_candidate.candidate_name #=> String
# resp.best_candidate.final_auto_ml_job_objective_metric.type #=> String, one of "Maximize", "Minimize"
# resp.best_candidate.final_auto_ml_job_objective_metric.metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.best_candidate.final_auto_ml_job_objective_metric.value #=> Float
# resp.best_candidate.final_auto_ml_job_objective_metric.standard_metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.best_candidate.objective_status #=> String, one of "Succeeded", "Pending", "Failed"
# resp.best_candidate.candidate_steps #=> Array
# resp.best_candidate.candidate_steps[0].candidate_step_type #=> String, one of "AWS::SageMaker::TrainingJob", "AWS::SageMaker::TransformJob", "AWS::SageMaker::ProcessingJob"
# resp.best_candidate.candidate_steps[0].candidate_step_arn #=> String
# resp.best_candidate.candidate_steps[0].candidate_step_name #=> String
# resp.best_candidate.candidate_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.best_candidate.inference_containers #=> Array
# resp.best_candidate.inference_containers[0].image #=> String
# resp.best_candidate.inference_containers[0].model_data_url #=> String
# resp.best_candidate.inference_containers[0].environment #=> Hash
# resp.best_candidate.inference_containers[0].environment["EnvironmentKey"] #=> String
# resp.best_candidate.creation_time #=> Time
# resp.best_candidate.end_time #=> Time
# resp.best_candidate.last_modified_time #=> Time
# resp.best_candidate.failure_reason #=> String
# resp.best_candidate.candidate_properties.candidate_artifact_locations.explainability #=> String
# resp.best_candidate.candidate_properties.candidate_artifact_locations.model_insights #=> String
# resp.best_candidate.candidate_properties.candidate_metrics #=> Array
# resp.best_candidate.candidate_properties.candidate_metrics[0].metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.best_candidate.candidate_properties.candidate_metrics[0].value #=> Float
# resp.best_candidate.candidate_properties.candidate_metrics[0].set #=> String, one of "Train", "Validation", "Test"
# resp.best_candidate.candidate_properties.candidate_metrics[0].standard_metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro", "LogLoss", "InferenceLatency"
# resp.best_candidate.inference_container_definitions #=> Hash
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"] #=> Array
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].image #=> String
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].model_data_url #=> String
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].environment #=> Hash
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].environment["EnvironmentKey"] #=> String
# resp.auto_ml_job_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.auto_ml_job_secondary_status #=> String, one of "Starting", "AnalyzingData", "FeatureEngineering", "ModelTuning", "MaxCandidatesReached", "Failed", "Stopped", "MaxAutoMLJobRuntimeReached", "Stopping", "CandidateDefinitionsGenerated", "GeneratingExplainabilityReport", "Completed", "ExplainabilityError", "DeployingModel", "ModelDeploymentError", "GeneratingModelInsightsReport", "ModelInsightsError", "TrainingModels"
# resp.generate_candidate_definitions_only #=> Boolean
# resp.auto_ml_job_artifacts.candidate_definition_notebook_location #=> String
# resp.auto_ml_job_artifacts.data_exploration_notebook_location #=> String
# resp.resolved_attributes.auto_ml_job_objective.metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.resolved_attributes.problem_type #=> String, one of "BinaryClassification", "MulticlassClassification", "Regression"
# resp.resolved_attributes.completion_criteria.max_candidates #=> Integer
# resp.resolved_attributes.completion_criteria.max_runtime_per_training_job_in_seconds #=> Integer
# resp.resolved_attributes.completion_criteria.max_auto_ml_job_runtime_in_seconds #=> Integer
# resp.model_deploy_config.auto_generate_endpoint_name #=> Boolean
# resp.model_deploy_config.endpoint_name #=> String
# resp.model_deploy_result.endpoint_name #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeAutoMLJob AWS API Documentation
#
# @overload describe_auto_ml_job(params = {})
# @param [Hash] params ({})
def describe_auto_ml_job(params = {}, options = {})
req = build_request(:describe_auto_ml_job, params)
req.send_request(options)
end
# Returns information about an Amazon SageMaker AutoML V2 job.
#
# This API action is callable through SageMaker Canvas only. Calling it
# directly from the CLI or an SDK results in an error.
#
#
#
# @option params [required, String] :auto_ml_job_name
# Requests information about an AutoML V2 job using its unique name.
#
# @return [Types::DescribeAutoMLJobV2Response] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeAutoMLJobV2Response#auto_ml_job_name #auto_ml_job_name} => String
# * {Types::DescribeAutoMLJobV2Response#auto_ml_job_arn #auto_ml_job_arn} => String
# * {Types::DescribeAutoMLJobV2Response#auto_ml_job_input_data_config #auto_ml_job_input_data_config} => Array<Types::AutoMLJobChannel>
# * {Types::DescribeAutoMLJobV2Response#output_data_config #output_data_config} => Types::AutoMLOutputDataConfig
# * {Types::DescribeAutoMLJobV2Response#role_arn #role_arn} => String
# * {Types::DescribeAutoMLJobV2Response#auto_ml_job_objective #auto_ml_job_objective} => Types::AutoMLJobObjective
# * {Types::DescribeAutoMLJobV2Response#auto_ml_problem_type_config #auto_ml_problem_type_config} => Types::AutoMLProblemTypeConfig
# * {Types::DescribeAutoMLJobV2Response#creation_time #creation_time} => Time
# * {Types::DescribeAutoMLJobV2Response#end_time #end_time} => Time
# * {Types::DescribeAutoMLJobV2Response#last_modified_time #last_modified_time} => Time
# * {Types::DescribeAutoMLJobV2Response#failure_reason #failure_reason} => String
# * {Types::DescribeAutoMLJobV2Response#partial_failure_reasons #partial_failure_reasons} => Array<Types::AutoMLPartialFailureReason>
# * {Types::DescribeAutoMLJobV2Response#best_candidate #best_candidate} => Types::AutoMLCandidate
# * {Types::DescribeAutoMLJobV2Response#auto_ml_job_status #auto_ml_job_status} => String
# * {Types::DescribeAutoMLJobV2Response#auto_ml_job_secondary_status #auto_ml_job_secondary_status} => String
# * {Types::DescribeAutoMLJobV2Response#model_deploy_config #model_deploy_config} => Types::ModelDeployConfig
# * {Types::DescribeAutoMLJobV2Response#model_deploy_result #model_deploy_result} => Types::ModelDeployResult
# * {Types::DescribeAutoMLJobV2Response#data_split_config #data_split_config} => Types::AutoMLDataSplitConfig
# * {Types::DescribeAutoMLJobV2Response#security_config #security_config} => Types::AutoMLSecurityConfig
#
# @example Request syntax with placeholder values
#
# resp = client.describe_auto_ml_job_v2({
# auto_ml_job_name: "AutoMLJobName", # required
# })
#
# @example Response structure
#
# resp.auto_ml_job_name #=> String
# resp.auto_ml_job_arn #=> String
# resp.auto_ml_job_input_data_config #=> Array
# resp.auto_ml_job_input_data_config[0].channel_type #=> String, one of "training", "validation"
# resp.auto_ml_job_input_data_config[0].content_type #=> String
# resp.auto_ml_job_input_data_config[0].compression_type #=> String, one of "None", "Gzip"
# resp.auto_ml_job_input_data_config[0].data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.auto_ml_job_input_data_config[0].data_source.s3_data_source.s3_uri #=> String
# resp.output_data_config.kms_key_id #=> String
# resp.output_data_config.s3_output_path #=> String
# resp.role_arn #=> String
# resp.auto_ml_job_objective.metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.auto_ml_problem_type_config.image_classification_job_config.completion_criteria.max_candidates #=> Integer
# resp.auto_ml_problem_type_config.image_classification_job_config.completion_criteria.max_runtime_per_training_job_in_seconds #=> Integer
# resp.auto_ml_problem_type_config.image_classification_job_config.completion_criteria.max_auto_ml_job_runtime_in_seconds #=> Integer
# resp.auto_ml_problem_type_config.text_classification_job_config.completion_criteria.max_candidates #=> Integer
# resp.auto_ml_problem_type_config.text_classification_job_config.completion_criteria.max_runtime_per_training_job_in_seconds #=> Integer
# resp.auto_ml_problem_type_config.text_classification_job_config.completion_criteria.max_auto_ml_job_runtime_in_seconds #=> Integer
# resp.auto_ml_problem_type_config.text_classification_job_config.content_column #=> String
# resp.auto_ml_problem_type_config.text_classification_job_config.target_label_column #=> String
# resp.creation_time #=> Time
# resp.end_time #=> Time
# resp.last_modified_time #=> Time
# resp.failure_reason #=> String
# resp.partial_failure_reasons #=> Array
# resp.partial_failure_reasons[0].partial_failure_message #=> String
# resp.best_candidate.candidate_name #=> String
# resp.best_candidate.final_auto_ml_job_objective_metric.type #=> String, one of "Maximize", "Minimize"
# resp.best_candidate.final_auto_ml_job_objective_metric.metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.best_candidate.final_auto_ml_job_objective_metric.value #=> Float
# resp.best_candidate.final_auto_ml_job_objective_metric.standard_metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.best_candidate.objective_status #=> String, one of "Succeeded", "Pending", "Failed"
# resp.best_candidate.candidate_steps #=> Array
# resp.best_candidate.candidate_steps[0].candidate_step_type #=> String, one of "AWS::SageMaker::TrainingJob", "AWS::SageMaker::TransformJob", "AWS::SageMaker::ProcessingJob"
# resp.best_candidate.candidate_steps[0].candidate_step_arn #=> String
# resp.best_candidate.candidate_steps[0].candidate_step_name #=> String
# resp.best_candidate.candidate_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.best_candidate.inference_containers #=> Array
# resp.best_candidate.inference_containers[0].image #=> String
# resp.best_candidate.inference_containers[0].model_data_url #=> String
# resp.best_candidate.inference_containers[0].environment #=> Hash
# resp.best_candidate.inference_containers[0].environment["EnvironmentKey"] #=> String
# resp.best_candidate.creation_time #=> Time
# resp.best_candidate.end_time #=> Time
# resp.best_candidate.last_modified_time #=> Time
# resp.best_candidate.failure_reason #=> String
# resp.best_candidate.candidate_properties.candidate_artifact_locations.explainability #=> String
# resp.best_candidate.candidate_properties.candidate_artifact_locations.model_insights #=> String
# resp.best_candidate.candidate_properties.candidate_metrics #=> Array
# resp.best_candidate.candidate_properties.candidate_metrics[0].metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.best_candidate.candidate_properties.candidate_metrics[0].value #=> Float
# resp.best_candidate.candidate_properties.candidate_metrics[0].set #=> String, one of "Train", "Validation", "Test"
# resp.best_candidate.candidate_properties.candidate_metrics[0].standard_metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro", "LogLoss", "InferenceLatency"
# resp.best_candidate.inference_container_definitions #=> Hash
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"] #=> Array
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].image #=> String
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].model_data_url #=> String
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].environment #=> Hash
# resp.best_candidate.inference_container_definitions["AutoMLProcessingUnit"][0].environment["EnvironmentKey"] #=> String
# resp.auto_ml_job_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.auto_ml_job_secondary_status #=> String, one of "Starting", "AnalyzingData", "FeatureEngineering", "ModelTuning", "MaxCandidatesReached", "Failed", "Stopped", "MaxAutoMLJobRuntimeReached", "Stopping", "CandidateDefinitionsGenerated", "GeneratingExplainabilityReport", "Completed", "ExplainabilityError", "DeployingModel", "ModelDeploymentError", "GeneratingModelInsightsReport", "ModelInsightsError", "TrainingModels"
# resp.model_deploy_config.auto_generate_endpoint_name #=> Boolean
# resp.model_deploy_config.endpoint_name #=> String
# resp.model_deploy_result.endpoint_name #=> String
# resp.data_split_config.validation_fraction #=> Float
# resp.security_config.volume_kms_key_id #=> String
# resp.security_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.security_config.vpc_config.security_group_ids #=> Array
# resp.security_config.vpc_config.security_group_ids[0] #=> String
# resp.security_config.vpc_config.subnets #=> Array
# resp.security_config.vpc_config.subnets[0] #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeAutoMLJobV2 AWS API Documentation
#
# @overload describe_auto_ml_job_v2(params = {})
# @param [Hash] params ({})
def describe_auto_ml_job_v2(params = {}, options = {})
req = build_request(:describe_auto_ml_job_v2, params)
req.send_request(options)
end
# Gets details about the specified Git repository.
#
# @option params [required, String] :code_repository_name
# The name of the Git repository to describe.
#
# @return [Types::DescribeCodeRepositoryOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeCodeRepositoryOutput#code_repository_name #code_repository_name} => String
# * {Types::DescribeCodeRepositoryOutput#code_repository_arn #code_repository_arn} => String
# * {Types::DescribeCodeRepositoryOutput#creation_time #creation_time} => Time
# * {Types::DescribeCodeRepositoryOutput#last_modified_time #last_modified_time} => Time
# * {Types::DescribeCodeRepositoryOutput#git_config #git_config} => Types::GitConfig
#
# @example Request syntax with placeholder values
#
# resp = client.describe_code_repository({
# code_repository_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.code_repository_name #=> String
# resp.code_repository_arn #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.git_config.repository_url #=> String
# resp.git_config.branch #=> String
# resp.git_config.secret_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeCodeRepository AWS API Documentation
#
# @overload describe_code_repository(params = {})
# @param [Hash] params ({})
def describe_code_repository(params = {}, options = {})
req = build_request(:describe_code_repository, params)
req.send_request(options)
end
# Returns information about a model compilation job.
#
# To create a model compilation job, use [CreateCompilationJob][1]. To
# get information about multiple model compilation jobs, use
# [ListCompilationJobs][2].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateCompilationJob.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ListCompilationJobs.html
#
# @option params [required, String] :compilation_job_name
# The name of the model compilation job that you want information about.
#
# @return [Types::DescribeCompilationJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeCompilationJobResponse#compilation_job_name #compilation_job_name} => String
# * {Types::DescribeCompilationJobResponse#compilation_job_arn #compilation_job_arn} => String
# * {Types::DescribeCompilationJobResponse#compilation_job_status #compilation_job_status} => String
# * {Types::DescribeCompilationJobResponse#compilation_start_time #compilation_start_time} => Time
# * {Types::DescribeCompilationJobResponse#compilation_end_time #compilation_end_time} => Time
# * {Types::DescribeCompilationJobResponse#stopping_condition #stopping_condition} => Types::StoppingCondition
# * {Types::DescribeCompilationJobResponse#inference_image #inference_image} => String
# * {Types::DescribeCompilationJobResponse#model_package_version_arn #model_package_version_arn} => String
# * {Types::DescribeCompilationJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeCompilationJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeCompilationJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeCompilationJobResponse#model_artifacts #model_artifacts} => Types::ModelArtifacts
# * {Types::DescribeCompilationJobResponse#model_digests #model_digests} => Types::ModelDigests
# * {Types::DescribeCompilationJobResponse#role_arn #role_arn} => String
# * {Types::DescribeCompilationJobResponse#input_config #input_config} => Types::InputConfig
# * {Types::DescribeCompilationJobResponse#output_config #output_config} => Types::OutputConfig
# * {Types::DescribeCompilationJobResponse#vpc_config #vpc_config} => Types::NeoVpcConfig
#
# @example Request syntax with placeholder values
#
# resp = client.describe_compilation_job({
# compilation_job_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.compilation_job_name #=> String
# resp.compilation_job_arn #=> String
# resp.compilation_job_status #=> String, one of "INPROGRESS", "COMPLETED", "FAILED", "STARTING", "STOPPING", "STOPPED"
# resp.compilation_start_time #=> Time
# resp.compilation_end_time #=> Time
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
# resp.stopping_condition.max_wait_time_in_seconds #=> Integer
# resp.inference_image #=> String
# resp.model_package_version_arn #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.failure_reason #=> String
# resp.model_artifacts.s3_model_artifacts #=> String
# resp.model_digests.artifact_digest #=> String
# resp.role_arn #=> String
# resp.input_config.s3_uri #=> String
# resp.input_config.data_input_config #=> String
# resp.input_config.framework #=> String, one of "TENSORFLOW", "KERAS", "MXNET", "ONNX", "PYTORCH", "XGBOOST", "TFLITE", "DARKNET", "SKLEARN"
# resp.input_config.framework_version #=> String
# resp.output_config.s3_output_location #=> String
# resp.output_config.target_device #=> String, one of "lambda", "ml_m4", "ml_m5", "ml_c4", "ml_c5", "ml_p2", "ml_p3", "ml_g4dn", "ml_inf1", "ml_eia2", "jetson_tx1", "jetson_tx2", "jetson_nano", "jetson_xavier", "rasp3b", "imx8qm", "deeplens", "rk3399", "rk3288", "aisage", "sbe_c", "qcs605", "qcs603", "sitara_am57x", "amba_cv2", "amba_cv22", "amba_cv25", "x86_win32", "x86_win64", "coreml", "jacinto_tda4vm", "imx8mplus"
# resp.output_config.target_platform.os #=> String, one of "ANDROID", "LINUX"
# resp.output_config.target_platform.arch #=> String, one of "X86_64", "X86", "ARM64", "ARM_EABI", "ARM_EABIHF"
# resp.output_config.target_platform.accelerator #=> String, one of "INTEL_GRAPHICS", "MALI", "NVIDIA", "NNA"
# resp.output_config.compiler_options #=> String
# resp.output_config.kms_key_id #=> String
# resp.vpc_config.security_group_ids #=> Array
# resp.vpc_config.security_group_ids[0] #=> String
# resp.vpc_config.subnets #=> Array
# resp.vpc_config.subnets[0] #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeCompilationJob AWS API Documentation
#
# @overload describe_compilation_job(params = {})
# @param [Hash] params ({})
def describe_compilation_job(params = {}, options = {})
req = build_request(:describe_compilation_job, params)
req.send_request(options)
end
# Describes a context.
#
# @option params [required, String] :context_name
# The name of the context to describe.
#
# @return [Types::DescribeContextResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeContextResponse#context_name #context_name} => String
# * {Types::DescribeContextResponse#context_arn #context_arn} => String
# * {Types::DescribeContextResponse#source #source} => Types::ContextSource
# * {Types::DescribeContextResponse#context_type #context_type} => String
# * {Types::DescribeContextResponse#description #description} => String
# * {Types::DescribeContextResponse#properties #properties} => Hash<String,String>
# * {Types::DescribeContextResponse#creation_time #creation_time} => Time
# * {Types::DescribeContextResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeContextResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeContextResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeContextResponse#lineage_group_arn #lineage_group_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_context({
# context_name: "ExperimentEntityNameOrArn", # required
# })
#
# @example Response structure
#
# resp.context_name #=> String
# resp.context_arn #=> String
# resp.source.source_uri #=> String
# resp.source.source_type #=> String
# resp.source.source_id #=> String
# resp.context_type #=> String
# resp.description #=> String
# resp.properties #=> Hash
# resp.properties["StringParameterValue"] #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.lineage_group_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeContext AWS API Documentation
#
# @overload describe_context(params = {})
# @param [Hash] params ({})
def describe_context(params = {}, options = {})
req = build_request(:describe_context, params)
req.send_request(options)
end
# Gets the details of a data quality monitoring job definition.
#
# @option params [required, String] :job_definition_name
# The name of the data quality monitoring job definition to describe.
#
# @return [Types::DescribeDataQualityJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeDataQualityJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
# * {Types::DescribeDataQualityJobDefinitionResponse#job_definition_name #job_definition_name} => String
# * {Types::DescribeDataQualityJobDefinitionResponse#creation_time #creation_time} => Time
# * {Types::DescribeDataQualityJobDefinitionResponse#data_quality_baseline_config #data_quality_baseline_config} => Types::DataQualityBaselineConfig
# * {Types::DescribeDataQualityJobDefinitionResponse#data_quality_app_specification #data_quality_app_specification} => Types::DataQualityAppSpecification
# * {Types::DescribeDataQualityJobDefinitionResponse#data_quality_job_input #data_quality_job_input} => Types::DataQualityJobInput
# * {Types::DescribeDataQualityJobDefinitionResponse#data_quality_job_output_config #data_quality_job_output_config} => Types::MonitoringOutputConfig
# * {Types::DescribeDataQualityJobDefinitionResponse#job_resources #job_resources} => Types::MonitoringResources
# * {Types::DescribeDataQualityJobDefinitionResponse#network_config #network_config} => Types::MonitoringNetworkConfig
# * {Types::DescribeDataQualityJobDefinitionResponse#role_arn #role_arn} => String
# * {Types::DescribeDataQualityJobDefinitionResponse#stopping_condition #stopping_condition} => Types::MonitoringStoppingCondition
#
# @example Request syntax with placeholder values
#
# resp = client.describe_data_quality_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
# resp.job_definition_name #=> String
# resp.creation_time #=> Time
# resp.data_quality_baseline_config.baselining_job_name #=> String
# resp.data_quality_baseline_config.constraints_resource.s3_uri #=> String
# resp.data_quality_baseline_config.statistics_resource.s3_uri #=> String
# resp.data_quality_app_specification.image_uri #=> String
# resp.data_quality_app_specification.container_entrypoint #=> Array
# resp.data_quality_app_specification.container_entrypoint[0] #=> String
# resp.data_quality_app_specification.container_arguments #=> Array
# resp.data_quality_app_specification.container_arguments[0] #=> String
# resp.data_quality_app_specification.record_preprocessor_source_uri #=> String
# resp.data_quality_app_specification.post_analytics_processor_source_uri #=> String
# resp.data_quality_app_specification.environment #=> Hash
# resp.data_quality_app_specification.environment["ProcessingEnvironmentKey"] #=> String
# resp.data_quality_job_input.endpoint_input.endpoint_name #=> String
# resp.data_quality_job_input.endpoint_input.local_path #=> String
# resp.data_quality_job_input.endpoint_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.data_quality_job_input.endpoint_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.data_quality_job_input.endpoint_input.features_attribute #=> String
# resp.data_quality_job_input.endpoint_input.inference_attribute #=> String
# resp.data_quality_job_input.endpoint_input.probability_attribute #=> String
# resp.data_quality_job_input.endpoint_input.probability_threshold_attribute #=> Float
# resp.data_quality_job_input.endpoint_input.start_time_offset #=> String
# resp.data_quality_job_input.endpoint_input.end_time_offset #=> String
# resp.data_quality_job_input.batch_transform_input.data_captured_destination_s3_uri #=> String
# resp.data_quality_job_input.batch_transform_input.dataset_format.csv.header #=> Boolean
# resp.data_quality_job_input.batch_transform_input.dataset_format.json.line #=> Boolean
# resp.data_quality_job_input.batch_transform_input.local_path #=> String
# resp.data_quality_job_input.batch_transform_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.data_quality_job_input.batch_transform_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.data_quality_job_input.batch_transform_input.features_attribute #=> String
# resp.data_quality_job_input.batch_transform_input.inference_attribute #=> String
# resp.data_quality_job_input.batch_transform_input.probability_attribute #=> String
# resp.data_quality_job_input.batch_transform_input.probability_threshold_attribute #=> Float
# resp.data_quality_job_input.batch_transform_input.start_time_offset #=> String
# resp.data_quality_job_input.batch_transform_input.end_time_offset #=> String
# resp.data_quality_job_output_config.monitoring_outputs #=> Array
# resp.data_quality_job_output_config.monitoring_outputs[0].s3_output.s3_uri #=> String
# resp.data_quality_job_output_config.monitoring_outputs[0].s3_output.local_path #=> String
# resp.data_quality_job_output_config.monitoring_outputs[0].s3_output.s3_upload_mode #=> String, one of "Continuous", "EndOfJob"
# resp.data_quality_job_output_config.kms_key_id #=> String
# resp.job_resources.cluster_config.instance_count #=> Integer
# resp.job_resources.cluster_config.instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.job_resources.cluster_config.volume_size_in_gb #=> Integer
# resp.job_resources.cluster_config.volume_kms_key_id #=> String
# resp.network_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.network_config.enable_network_isolation #=> Boolean
# resp.network_config.vpc_config.security_group_ids #=> Array
# resp.network_config.vpc_config.security_group_ids[0] #=> String
# resp.network_config.vpc_config.subnets #=> Array
# resp.network_config.vpc_config.subnets[0] #=> String
# resp.role_arn #=> String
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeDataQualityJobDefinition AWS API Documentation
#
# @overload describe_data_quality_job_definition(params = {})
# @param [Hash] params ({})
def describe_data_quality_job_definition(params = {}, options = {})
req = build_request(:describe_data_quality_job_definition, params)
req.send_request(options)
end
# Describes the device.
#
# @option params [String] :next_token
# Next token of device description.
#
# @option params [required, String] :device_name
# The unique ID of the device.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet the devices belong to.
#
# @return [Types::DescribeDeviceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeDeviceResponse#device_arn #device_arn} => String
# * {Types::DescribeDeviceResponse#device_name #device_name} => String
# * {Types::DescribeDeviceResponse#description #description} => String
# * {Types::DescribeDeviceResponse#device_fleet_name #device_fleet_name} => String
# * {Types::DescribeDeviceResponse#iot_thing_name #iot_thing_name} => String
# * {Types::DescribeDeviceResponse#registration_time #registration_time} => Time
# * {Types::DescribeDeviceResponse#latest_heartbeat #latest_heartbeat} => Time
# * {Types::DescribeDeviceResponse#models #models} => Array<Types::EdgeModel>
# * {Types::DescribeDeviceResponse#max_models #max_models} => Integer
# * {Types::DescribeDeviceResponse#next_token #next_token} => String
# * {Types::DescribeDeviceResponse#agent_version #agent_version} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_device({
# next_token: "NextToken",
# device_name: "EntityName", # required
# device_fleet_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.device_arn #=> String
# resp.device_name #=> String
# resp.description #=> String
# resp.device_fleet_name #=> String
# resp.iot_thing_name #=> String
# resp.registration_time #=> Time
# resp.latest_heartbeat #=> Time
# resp.models #=> Array
# resp.models[0].model_name #=> String
# resp.models[0].model_version #=> String
# resp.models[0].latest_sample_time #=> Time
# resp.models[0].latest_inference #=> Time
# resp.max_models #=> Integer
# resp.next_token #=> String
# resp.agent_version #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeDevice AWS API Documentation
#
# @overload describe_device(params = {})
# @param [Hash] params ({})
def describe_device(params = {}, options = {})
req = build_request(:describe_device, params)
req.send_request(options)
end
# A description of the fleet the device belongs to.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet.
#
# @return [Types::DescribeDeviceFleetResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeDeviceFleetResponse#device_fleet_name #device_fleet_name} => String
# * {Types::DescribeDeviceFleetResponse#device_fleet_arn #device_fleet_arn} => String
# * {Types::DescribeDeviceFleetResponse#output_config #output_config} => Types::EdgeOutputConfig
# * {Types::DescribeDeviceFleetResponse#description #description} => String
# * {Types::DescribeDeviceFleetResponse#creation_time #creation_time} => Time
# * {Types::DescribeDeviceFleetResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeDeviceFleetResponse#role_arn #role_arn} => String
# * {Types::DescribeDeviceFleetResponse#iot_role_alias #iot_role_alias} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_device_fleet({
# device_fleet_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.device_fleet_name #=> String
# resp.device_fleet_arn #=> String
# resp.output_config.s3_output_location #=> String
# resp.output_config.kms_key_id #=> String
# resp.output_config.preset_deployment_type #=> String, one of "GreengrassV2Component"
# resp.output_config.preset_deployment_config #=> String
# resp.description #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.role_arn #=> String
# resp.iot_role_alias #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeDeviceFleet AWS API Documentation
#
# @overload describe_device_fleet(params = {})
# @param [Hash] params ({})
def describe_device_fleet(params = {}, options = {})
req = build_request(:describe_device_fleet, params)
req.send_request(options)
end
# The description of the domain.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @return [Types::DescribeDomainResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeDomainResponse#domain_arn #domain_arn} => String
# * {Types::DescribeDomainResponse#domain_id #domain_id} => String
# * {Types::DescribeDomainResponse#domain_name #domain_name} => String
# * {Types::DescribeDomainResponse#home_efs_file_system_id #home_efs_file_system_id} => String
# * {Types::DescribeDomainResponse#single_sign_on_managed_application_instance_id #single_sign_on_managed_application_instance_id} => String
# * {Types::DescribeDomainResponse#status #status} => String
# * {Types::DescribeDomainResponse#creation_time #creation_time} => Time
# * {Types::DescribeDomainResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeDomainResponse#failure_reason #failure_reason} => String
# * {Types::DescribeDomainResponse#auth_mode #auth_mode} => String
# * {Types::DescribeDomainResponse#default_user_settings #default_user_settings} => Types::UserSettings
# * {Types::DescribeDomainResponse#app_network_access_type #app_network_access_type} => String
# * {Types::DescribeDomainResponse#home_efs_file_system_kms_key_id #home_efs_file_system_kms_key_id} => String
# * {Types::DescribeDomainResponse#subnet_ids #subnet_ids} => Array<String>
# * {Types::DescribeDomainResponse#url #url} => String
# * {Types::DescribeDomainResponse#vpc_id #vpc_id} => String
# * {Types::DescribeDomainResponse#kms_key_id #kms_key_id} => String
# * {Types::DescribeDomainResponse#domain_settings #domain_settings} => Types::DomainSettings
# * {Types::DescribeDomainResponse#app_security_group_management #app_security_group_management} => String
# * {Types::DescribeDomainResponse#security_group_id_for_domain_boundary #security_group_id_for_domain_boundary} => String
# * {Types::DescribeDomainResponse#default_space_settings #default_space_settings} => Types::DefaultSpaceSettings
#
# @example Request syntax with placeholder values
#
# resp = client.describe_domain({
# domain_id: "DomainId", # required
# })
#
# @example Response structure
#
# resp.domain_arn #=> String
# resp.domain_id #=> String
# resp.domain_name #=> String
# resp.home_efs_file_system_id #=> String
# resp.single_sign_on_managed_application_instance_id #=> String
# resp.status #=> String, one of "Deleting", "Failed", "InService", "Pending", "Updating", "Update_Failed", "Delete_Failed"
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.failure_reason #=> String
# resp.auth_mode #=> String, one of "SSO", "IAM"
# resp.default_user_settings.execution_role #=> String
# resp.default_user_settings.security_groups #=> Array
# resp.default_user_settings.security_groups[0] #=> String
# resp.default_user_settings.sharing_settings.notebook_output_option #=> String, one of "Allowed", "Disabled"
# resp.default_user_settings.sharing_settings.s3_output_path #=> String
# resp.default_user_settings.sharing_settings.s3_kms_key_id #=> String
# resp.default_user_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.default_user_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.default_user_settings.jupyter_server_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.default_user_settings.jupyter_server_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.default_user_settings.jupyter_server_app_settings.lifecycle_config_arns #=> Array
# resp.default_user_settings.jupyter_server_app_settings.lifecycle_config_arns[0] #=> String
# resp.default_user_settings.jupyter_server_app_settings.code_repositories #=> Array
# resp.default_user_settings.jupyter_server_app_settings.code_repositories[0].repository_url #=> String
# resp.default_user_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.default_user_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.default_user_settings.kernel_gateway_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.default_user_settings.kernel_gateway_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.default_user_settings.kernel_gateway_app_settings.custom_images #=> Array
# resp.default_user_settings.kernel_gateway_app_settings.custom_images[0].image_name #=> String
# resp.default_user_settings.kernel_gateway_app_settings.custom_images[0].image_version_number #=> Integer
# resp.default_user_settings.kernel_gateway_app_settings.custom_images[0].app_image_config_name #=> String
# resp.default_user_settings.kernel_gateway_app_settings.lifecycle_config_arns #=> Array
# resp.default_user_settings.kernel_gateway_app_settings.lifecycle_config_arns[0] #=> String
# resp.default_user_settings.tensor_board_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.default_user_settings.tensor_board_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.default_user_settings.tensor_board_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.default_user_settings.tensor_board_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.default_user_settings.r_studio_server_pro_app_settings.access_status #=> String, one of "ENABLED", "DISABLED"
# resp.default_user_settings.r_studio_server_pro_app_settings.user_group #=> String, one of "R_STUDIO_ADMIN", "R_STUDIO_USER"
# resp.default_user_settings.r_session_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.default_user_settings.r_session_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.default_user_settings.r_session_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.default_user_settings.r_session_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.default_user_settings.r_session_app_settings.custom_images #=> Array
# resp.default_user_settings.r_session_app_settings.custom_images[0].image_name #=> String
# resp.default_user_settings.r_session_app_settings.custom_images[0].image_version_number #=> Integer
# resp.default_user_settings.r_session_app_settings.custom_images[0].app_image_config_name #=> String
# resp.default_user_settings.canvas_app_settings.time_series_forecasting_settings.status #=> String, one of "ENABLED", "DISABLED"
# resp.default_user_settings.canvas_app_settings.time_series_forecasting_settings.amazon_forecast_role_arn #=> String
# resp.default_user_settings.canvas_app_settings.model_register_settings.status #=> String, one of "ENABLED", "DISABLED"
# resp.default_user_settings.canvas_app_settings.model_register_settings.cross_account_model_register_role_arn #=> String
# resp.app_network_access_type #=> String, one of "PublicInternetOnly", "VpcOnly"
# resp.home_efs_file_system_kms_key_id #=> String
# resp.subnet_ids #=> Array
# resp.subnet_ids[0] #=> String
# resp.url #=> String
# resp.vpc_id #=> String
# resp.kms_key_id #=> String
# resp.domain_settings.security_group_ids #=> Array
# resp.domain_settings.security_group_ids[0] #=> String
# resp.domain_settings.r_studio_server_pro_domain_settings.domain_execution_role_arn #=> String
# resp.domain_settings.r_studio_server_pro_domain_settings.r_studio_connect_url #=> String
# resp.domain_settings.r_studio_server_pro_domain_settings.r_studio_package_manager_url #=> String
# resp.domain_settings.r_studio_server_pro_domain_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.domain_settings.r_studio_server_pro_domain_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.domain_settings.r_studio_server_pro_domain_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.domain_settings.r_studio_server_pro_domain_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.domain_settings.execution_role_identity_config #=> String, one of "USER_PROFILE_NAME", "DISABLED"
# resp.app_security_group_management #=> String, one of "Service", "Customer"
# resp.security_group_id_for_domain_boundary #=> String
# resp.default_space_settings.execution_role #=> String
# resp.default_space_settings.security_groups #=> Array
# resp.default_space_settings.security_groups[0] #=> String
# resp.default_space_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.default_space_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.default_space_settings.jupyter_server_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.default_space_settings.jupyter_server_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.default_space_settings.jupyter_server_app_settings.lifecycle_config_arns #=> Array
# resp.default_space_settings.jupyter_server_app_settings.lifecycle_config_arns[0] #=> String
# resp.default_space_settings.jupyter_server_app_settings.code_repositories #=> Array
# resp.default_space_settings.jupyter_server_app_settings.code_repositories[0].repository_url #=> String
# resp.default_space_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.default_space_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.default_space_settings.kernel_gateway_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.default_space_settings.kernel_gateway_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.default_space_settings.kernel_gateway_app_settings.custom_images #=> Array
# resp.default_space_settings.kernel_gateway_app_settings.custom_images[0].image_name #=> String
# resp.default_space_settings.kernel_gateway_app_settings.custom_images[0].image_version_number #=> Integer
# resp.default_space_settings.kernel_gateway_app_settings.custom_images[0].app_image_config_name #=> String
# resp.default_space_settings.kernel_gateway_app_settings.lifecycle_config_arns #=> Array
# resp.default_space_settings.kernel_gateway_app_settings.lifecycle_config_arns[0] #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeDomain AWS API Documentation
#
# @overload describe_domain(params = {})
# @param [Hash] params ({})
def describe_domain(params = {}, options = {})
req = build_request(:describe_domain, params)
req.send_request(options)
end
# Describes an edge deployment plan with deployment status per stage.
#
# @option params [required, String] :edge_deployment_plan_name
# The name of the deployment plan to describe.
#
# @option params [String] :next_token
# If the edge deployment plan has enough stages to require tokening,
# then this is the response from the last list of stages returned.
#
# @option params [Integer] :max_results
# The maximum number of results to select (50 by default).
#
# @return [Types::DescribeEdgeDeploymentPlanResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeEdgeDeploymentPlanResponse#edge_deployment_plan_arn #edge_deployment_plan_arn} => String
# * {Types::DescribeEdgeDeploymentPlanResponse#edge_deployment_plan_name #edge_deployment_plan_name} => String
# * {Types::DescribeEdgeDeploymentPlanResponse#model_configs #model_configs} => Array<Types::EdgeDeploymentModelConfig>
# * {Types::DescribeEdgeDeploymentPlanResponse#device_fleet_name #device_fleet_name} => String
# * {Types::DescribeEdgeDeploymentPlanResponse#edge_deployment_success #edge_deployment_success} => Integer
# * {Types::DescribeEdgeDeploymentPlanResponse#edge_deployment_pending #edge_deployment_pending} => Integer
# * {Types::DescribeEdgeDeploymentPlanResponse#edge_deployment_failed #edge_deployment_failed} => Integer
# * {Types::DescribeEdgeDeploymentPlanResponse#stages #stages} => Array<Types::DeploymentStageStatusSummary>
# * {Types::DescribeEdgeDeploymentPlanResponse#next_token #next_token} => String
# * {Types::DescribeEdgeDeploymentPlanResponse#creation_time #creation_time} => Time
# * {Types::DescribeEdgeDeploymentPlanResponse#last_modified_time #last_modified_time} => Time
#
# @example Request syntax with placeholder values
#
# resp = client.describe_edge_deployment_plan({
# edge_deployment_plan_name: "EntityName", # required
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.edge_deployment_plan_arn #=> String
# resp.edge_deployment_plan_name #=> String
# resp.model_configs #=> Array
# resp.model_configs[0].model_handle #=> String
# resp.model_configs[0].edge_packaging_job_name #=> String
# resp.device_fleet_name #=> String
# resp.edge_deployment_success #=> Integer
# resp.edge_deployment_pending #=> Integer
# resp.edge_deployment_failed #=> Integer
# resp.stages #=> Array
# resp.stages[0].stage_name #=> String
# resp.stages[0].device_selection_config.device_subset_type #=> String, one of "PERCENTAGE", "SELECTION", "NAMECONTAINS"
# resp.stages[0].device_selection_config.percentage #=> Integer
# resp.stages[0].device_selection_config.device_names #=> Array
# resp.stages[0].device_selection_config.device_names[0] #=> String
# resp.stages[0].device_selection_config.device_name_contains #=> String
# resp.stages[0].deployment_config.failure_handling_policy #=> String, one of "ROLLBACK_ON_FAILURE", "DO_NOTHING"
# resp.stages[0].deployment_status.stage_status #=> String, one of "CREATING", "READYTODEPLOY", "STARTING", "INPROGRESS", "DEPLOYED", "FAILED", "STOPPING", "STOPPED"
# resp.stages[0].deployment_status.edge_deployment_success_in_stage #=> Integer
# resp.stages[0].deployment_status.edge_deployment_pending_in_stage #=> Integer
# resp.stages[0].deployment_status.edge_deployment_failed_in_stage #=> Integer
# resp.stages[0].deployment_status.edge_deployment_status_message #=> String
# resp.stages[0].deployment_status.edge_deployment_stage_start_time #=> Time
# resp.next_token #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeEdgeDeploymentPlan AWS API Documentation
#
# @overload describe_edge_deployment_plan(params = {})
# @param [Hash] params ({})
def describe_edge_deployment_plan(params = {}, options = {})
req = build_request(:describe_edge_deployment_plan, params)
req.send_request(options)
end
# A description of edge packaging jobs.
#
# @option params [required, String] :edge_packaging_job_name
# The name of the edge packaging job.
#
# @return [Types::DescribeEdgePackagingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeEdgePackagingJobResponse#edge_packaging_job_arn #edge_packaging_job_arn} => String
# * {Types::DescribeEdgePackagingJobResponse#edge_packaging_job_name #edge_packaging_job_name} => String
# * {Types::DescribeEdgePackagingJobResponse#compilation_job_name #compilation_job_name} => String
# * {Types::DescribeEdgePackagingJobResponse#model_name #model_name} => String
# * {Types::DescribeEdgePackagingJobResponse#model_version #model_version} => String
# * {Types::DescribeEdgePackagingJobResponse#role_arn #role_arn} => String
# * {Types::DescribeEdgePackagingJobResponse#output_config #output_config} => Types::EdgeOutputConfig
# * {Types::DescribeEdgePackagingJobResponse#resource_key #resource_key} => String
# * {Types::DescribeEdgePackagingJobResponse#edge_packaging_job_status #edge_packaging_job_status} => String
# * {Types::DescribeEdgePackagingJobResponse#edge_packaging_job_status_message #edge_packaging_job_status_message} => String
# * {Types::DescribeEdgePackagingJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeEdgePackagingJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeEdgePackagingJobResponse#model_artifact #model_artifact} => String
# * {Types::DescribeEdgePackagingJobResponse#model_signature #model_signature} => String
# * {Types::DescribeEdgePackagingJobResponse#preset_deployment_output #preset_deployment_output} => Types::EdgePresetDeploymentOutput
#
# @example Request syntax with placeholder values
#
# resp = client.describe_edge_packaging_job({
# edge_packaging_job_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.edge_packaging_job_arn #=> String
# resp.edge_packaging_job_name #=> String
# resp.compilation_job_name #=> String
# resp.model_name #=> String
# resp.model_version #=> String
# resp.role_arn #=> String
# resp.output_config.s3_output_location #=> String
# resp.output_config.kms_key_id #=> String
# resp.output_config.preset_deployment_type #=> String, one of "GreengrassV2Component"
# resp.output_config.preset_deployment_config #=> String
# resp.resource_key #=> String
# resp.edge_packaging_job_status #=> String, one of "STARTING", "INPROGRESS", "COMPLETED", "FAILED", "STOPPING", "STOPPED"
# resp.edge_packaging_job_status_message #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.model_artifact #=> String
# resp.model_signature #=> String
# resp.preset_deployment_output.type #=> String, one of "GreengrassV2Component"
# resp.preset_deployment_output.artifact #=> String
# resp.preset_deployment_output.status #=> String, one of "COMPLETED", "FAILED"
# resp.preset_deployment_output.status_message #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeEdgePackagingJob AWS API Documentation
#
# @overload describe_edge_packaging_job(params = {})
# @param [Hash] params ({})
def describe_edge_packaging_job(params = {}, options = {})
req = build_request(:describe_edge_packaging_job, params)
req.send_request(options)
end
# Returns the description of an endpoint.
#
# @option params [required, String] :endpoint_name
# The name of the endpoint.
#
# @return [Types::DescribeEndpointOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeEndpointOutput#endpoint_name #endpoint_name} => String
# * {Types::DescribeEndpointOutput#endpoint_arn #endpoint_arn} => String
# * {Types::DescribeEndpointOutput#endpoint_config_name #endpoint_config_name} => String
# * {Types::DescribeEndpointOutput#production_variants #production_variants} => Array<Types::ProductionVariantSummary>
# * {Types::DescribeEndpointOutput#data_capture_config #data_capture_config} => Types::DataCaptureConfigSummary
# * {Types::DescribeEndpointOutput#endpoint_status #endpoint_status} => String
# * {Types::DescribeEndpointOutput#failure_reason #failure_reason} => String
# * {Types::DescribeEndpointOutput#creation_time #creation_time} => Time
# * {Types::DescribeEndpointOutput#last_modified_time #last_modified_time} => Time
# * {Types::DescribeEndpointOutput#last_deployment_config #last_deployment_config} => Types::DeploymentConfig
# * {Types::DescribeEndpointOutput#async_inference_config #async_inference_config} => Types::AsyncInferenceConfig
# * {Types::DescribeEndpointOutput#pending_deployment_summary #pending_deployment_summary} => Types::PendingDeploymentSummary
# * {Types::DescribeEndpointOutput#explainer_config #explainer_config} => Types::ExplainerConfig
# * {Types::DescribeEndpointOutput#shadow_production_variants #shadow_production_variants} => Array<Types::ProductionVariantSummary>
#
# @example Request syntax with placeholder values
#
# resp = client.describe_endpoint({
# endpoint_name: "EndpointName", # required
# })
#
# @example Response structure
#
# resp.endpoint_name #=> String
# resp.endpoint_arn #=> String
# resp.endpoint_config_name #=> String
# resp.production_variants #=> Array
# resp.production_variants[0].variant_name #=> String
# resp.production_variants[0].deployed_images #=> Array
# resp.production_variants[0].deployed_images[0].specified_image #=> String
# resp.production_variants[0].deployed_images[0].resolved_image #=> String
# resp.production_variants[0].deployed_images[0].resolution_time #=> Time
# resp.production_variants[0].current_weight #=> Float
# resp.production_variants[0].desired_weight #=> Float
# resp.production_variants[0].current_instance_count #=> Integer
# resp.production_variants[0].desired_instance_count #=> Integer
# resp.production_variants[0].variant_status #=> Array
# resp.production_variants[0].variant_status[0].status #=> String, one of "Creating", "Updating", "Deleting", "ActivatingTraffic", "Baking"
# resp.production_variants[0].variant_status[0].status_message #=> String
# resp.production_variants[0].variant_status[0].start_time #=> Time
# resp.production_variants[0].current_serverless_config.memory_size_in_mb #=> Integer
# resp.production_variants[0].current_serverless_config.max_concurrency #=> Integer
# resp.production_variants[0].desired_serverless_config.memory_size_in_mb #=> Integer
# resp.production_variants[0].desired_serverless_config.max_concurrency #=> Integer
# resp.data_capture_config.enable_capture #=> Boolean
# resp.data_capture_config.capture_status #=> String, one of "Started", "Stopped"
# resp.data_capture_config.current_sampling_percentage #=> Integer
# resp.data_capture_config.destination_s3_uri #=> String
# resp.data_capture_config.kms_key_id #=> String
# resp.endpoint_status #=> String, one of "OutOfService", "Creating", "Updating", "SystemUpdating", "RollingBack", "InService", "Deleting", "Failed"
# resp.failure_reason #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.last_deployment_config.blue_green_update_policy.traffic_routing_configuration.type #=> String, one of "ALL_AT_ONCE", "CANARY", "LINEAR"
# resp.last_deployment_config.blue_green_update_policy.traffic_routing_configuration.wait_interval_in_seconds #=> Integer
# resp.last_deployment_config.blue_green_update_policy.traffic_routing_configuration.canary_size.type #=> String, one of "INSTANCE_COUNT", "CAPACITY_PERCENT"
# resp.last_deployment_config.blue_green_update_policy.traffic_routing_configuration.canary_size.value #=> Integer
# resp.last_deployment_config.blue_green_update_policy.traffic_routing_configuration.linear_step_size.type #=> String, one of "INSTANCE_COUNT", "CAPACITY_PERCENT"
# resp.last_deployment_config.blue_green_update_policy.traffic_routing_configuration.linear_step_size.value #=> Integer
# resp.last_deployment_config.blue_green_update_policy.termination_wait_in_seconds #=> Integer
# resp.last_deployment_config.blue_green_update_policy.maximum_execution_timeout_in_seconds #=> Integer
# resp.last_deployment_config.auto_rollback_configuration.alarms #=> Array
# resp.last_deployment_config.auto_rollback_configuration.alarms[0].alarm_name #=> String
# resp.async_inference_config.client_config.max_concurrent_invocations_per_instance #=> Integer
# resp.async_inference_config.output_config.kms_key_id #=> String
# resp.async_inference_config.output_config.s3_output_path #=> String
# resp.async_inference_config.output_config.notification_config.success_topic #=> String
# resp.async_inference_config.output_config.notification_config.error_topic #=> String
# resp.async_inference_config.output_config.notification_config.include_inference_response_in #=> Array
# resp.async_inference_config.output_config.notification_config.include_inference_response_in[0] #=> String, one of "SUCCESS_NOTIFICATION_TOPIC", "ERROR_NOTIFICATION_TOPIC"
# resp.async_inference_config.output_config.s3_failure_path #=> String
# resp.pending_deployment_summary.endpoint_config_name #=> String
# resp.pending_deployment_summary.production_variants #=> Array
# resp.pending_deployment_summary.production_variants[0].variant_name #=> String
# resp.pending_deployment_summary.production_variants[0].deployed_images #=> Array
# resp.pending_deployment_summary.production_variants[0].deployed_images[0].specified_image #=> String
# resp.pending_deployment_summary.production_variants[0].deployed_images[0].resolved_image #=> String
# resp.pending_deployment_summary.production_variants[0].deployed_images[0].resolution_time #=> Time
# resp.pending_deployment_summary.production_variants[0].current_weight #=> Float
# resp.pending_deployment_summary.production_variants[0].desired_weight #=> Float
# resp.pending_deployment_summary.production_variants[0].current_instance_count #=> Integer
# resp.pending_deployment_summary.production_variants[0].desired_instance_count #=> Integer
# resp.pending_deployment_summary.production_variants[0].instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.pending_deployment_summary.production_variants[0].accelerator_type #=> String, one of "ml.eia1.medium", "ml.eia1.large", "ml.eia1.xlarge", "ml.eia2.medium", "ml.eia2.large", "ml.eia2.xlarge"
# resp.pending_deployment_summary.production_variants[0].variant_status #=> Array
# resp.pending_deployment_summary.production_variants[0].variant_status[0].status #=> String, one of "Creating", "Updating", "Deleting", "ActivatingTraffic", "Baking"
# resp.pending_deployment_summary.production_variants[0].variant_status[0].status_message #=> String
# resp.pending_deployment_summary.production_variants[0].variant_status[0].start_time #=> Time
# resp.pending_deployment_summary.production_variants[0].current_serverless_config.memory_size_in_mb #=> Integer
# resp.pending_deployment_summary.production_variants[0].current_serverless_config.max_concurrency #=> Integer
# resp.pending_deployment_summary.production_variants[0].desired_serverless_config.memory_size_in_mb #=> Integer
# resp.pending_deployment_summary.production_variants[0].desired_serverless_config.max_concurrency #=> Integer
# resp.pending_deployment_summary.start_time #=> Time
# resp.pending_deployment_summary.shadow_production_variants #=> Array
# resp.pending_deployment_summary.shadow_production_variants[0].variant_name #=> String
# resp.pending_deployment_summary.shadow_production_variants[0].deployed_images #=> Array
# resp.pending_deployment_summary.shadow_production_variants[0].deployed_images[0].specified_image #=> String
# resp.pending_deployment_summary.shadow_production_variants[0].deployed_images[0].resolved_image #=> String
# resp.pending_deployment_summary.shadow_production_variants[0].deployed_images[0].resolution_time #=> Time
# resp.pending_deployment_summary.shadow_production_variants[0].current_weight #=> Float
# resp.pending_deployment_summary.shadow_production_variants[0].desired_weight #=> Float
# resp.pending_deployment_summary.shadow_production_variants[0].current_instance_count #=> Integer
# resp.pending_deployment_summary.shadow_production_variants[0].desired_instance_count #=> Integer
# resp.pending_deployment_summary.shadow_production_variants[0].instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.pending_deployment_summary.shadow_production_variants[0].accelerator_type #=> String, one of "ml.eia1.medium", "ml.eia1.large", "ml.eia1.xlarge", "ml.eia2.medium", "ml.eia2.large", "ml.eia2.xlarge"
# resp.pending_deployment_summary.shadow_production_variants[0].variant_status #=> Array
# resp.pending_deployment_summary.shadow_production_variants[0].variant_status[0].status #=> String, one of "Creating", "Updating", "Deleting", "ActivatingTraffic", "Baking"
# resp.pending_deployment_summary.shadow_production_variants[0].variant_status[0].status_message #=> String
# resp.pending_deployment_summary.shadow_production_variants[0].variant_status[0].start_time #=> Time
# resp.pending_deployment_summary.shadow_production_variants[0].current_serverless_config.memory_size_in_mb #=> Integer
# resp.pending_deployment_summary.shadow_production_variants[0].current_serverless_config.max_concurrency #=> Integer
# resp.pending_deployment_summary.shadow_production_variants[0].desired_serverless_config.memory_size_in_mb #=> Integer
# resp.pending_deployment_summary.shadow_production_variants[0].desired_serverless_config.max_concurrency #=> Integer
# resp.explainer_config.clarify_explainer_config.enable_explanations #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.features_attribute #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.content_template #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.max_record_count #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.max_payload_in_mb #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.probability_index #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.label_index #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.probability_attribute #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.label_attribute #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.label_headers #=> Array
# resp.explainer_config.clarify_explainer_config.inference_config.label_headers[0] #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.feature_headers #=> Array
# resp.explainer_config.clarify_explainer_config.inference_config.feature_headers[0] #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.feature_types #=> Array
# resp.explainer_config.clarify_explainer_config.inference_config.feature_types[0] #=> String, one of "numerical", "categorical", "text"
# resp.explainer_config.clarify_explainer_config.shap_config.shap_baseline_config.mime_type #=> String
# resp.explainer_config.clarify_explainer_config.shap_config.shap_baseline_config.shap_baseline #=> String
# resp.explainer_config.clarify_explainer_config.shap_config.shap_baseline_config.shap_baseline_uri #=> String
# resp.explainer_config.clarify_explainer_config.shap_config.number_of_samples #=> Integer
# resp.explainer_config.clarify_explainer_config.shap_config.use_logit #=> Boolean
# resp.explainer_config.clarify_explainer_config.shap_config.seed #=> Integer
# resp.explainer_config.clarify_explainer_config.shap_config.text_config.language #=> String, one of "af", "sq", "ar", "hy", "eu", "bn", "bg", "ca", "zh", "hr", "cs", "da", "nl", "en", "et", "fi", "fr", "de", "el", "gu", "he", "hi", "hu", "is", "id", "ga", "it", "kn", "ky", "lv", "lt", "lb", "mk", "ml", "mr", "ne", "nb", "fa", "pl", "pt", "ro", "ru", "sa", "sr", "tn", "si", "sk", "sl", "es", "sv", "tl", "ta", "tt", "te", "tr", "uk", "ur", "yo", "lij", "xx"
# resp.explainer_config.clarify_explainer_config.shap_config.text_config.granularity #=> String, one of "token", "sentence", "paragraph"
# resp.shadow_production_variants #=> Array
# resp.shadow_production_variants[0].variant_name #=> String
# resp.shadow_production_variants[0].deployed_images #=> Array
# resp.shadow_production_variants[0].deployed_images[0].specified_image #=> String
# resp.shadow_production_variants[0].deployed_images[0].resolved_image #=> String
# resp.shadow_production_variants[0].deployed_images[0].resolution_time #=> Time
# resp.shadow_production_variants[0].current_weight #=> Float
# resp.shadow_production_variants[0].desired_weight #=> Float
# resp.shadow_production_variants[0].current_instance_count #=> Integer
# resp.shadow_production_variants[0].desired_instance_count #=> Integer
# resp.shadow_production_variants[0].variant_status #=> Array
# resp.shadow_production_variants[0].variant_status[0].status #=> String, one of "Creating", "Updating", "Deleting", "ActivatingTraffic", "Baking"
# resp.shadow_production_variants[0].variant_status[0].status_message #=> String
# resp.shadow_production_variants[0].variant_status[0].start_time #=> Time
# resp.shadow_production_variants[0].current_serverless_config.memory_size_in_mb #=> Integer
# resp.shadow_production_variants[0].current_serverless_config.max_concurrency #=> Integer
# resp.shadow_production_variants[0].desired_serverless_config.memory_size_in_mb #=> Integer
# resp.shadow_production_variants[0].desired_serverless_config.max_concurrency #=> Integer
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * endpoint_deleted
# * endpoint_in_service
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeEndpoint AWS API Documentation
#
# @overload describe_endpoint(params = {})
# @param [Hash] params ({})
def describe_endpoint(params = {}, options = {})
req = build_request(:describe_endpoint, params)
req.send_request(options)
end
# Returns the description of an endpoint configuration created using the
# `CreateEndpointConfig` API.
#
# @option params [required, String] :endpoint_config_name
# The name of the endpoint configuration.
#
# @return [Types::DescribeEndpointConfigOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeEndpointConfigOutput#endpoint_config_name #endpoint_config_name} => String
# * {Types::DescribeEndpointConfigOutput#endpoint_config_arn #endpoint_config_arn} => String
# * {Types::DescribeEndpointConfigOutput#production_variants #production_variants} => Array<Types::ProductionVariant>
# * {Types::DescribeEndpointConfigOutput#data_capture_config #data_capture_config} => Types::DataCaptureConfig
# * {Types::DescribeEndpointConfigOutput#kms_key_id #kms_key_id} => String
# * {Types::DescribeEndpointConfigOutput#creation_time #creation_time} => Time
# * {Types::DescribeEndpointConfigOutput#async_inference_config #async_inference_config} => Types::AsyncInferenceConfig
# * {Types::DescribeEndpointConfigOutput#explainer_config #explainer_config} => Types::ExplainerConfig
# * {Types::DescribeEndpointConfigOutput#shadow_production_variants #shadow_production_variants} => Array<Types::ProductionVariant>
#
# @example Request syntax with placeholder values
#
# resp = client.describe_endpoint_config({
# endpoint_config_name: "EndpointConfigName", # required
# })
#
# @example Response structure
#
# resp.endpoint_config_name #=> String
# resp.endpoint_config_arn #=> String
# resp.production_variants #=> Array
# resp.production_variants[0].variant_name #=> String
# resp.production_variants[0].model_name #=> String
# resp.production_variants[0].initial_instance_count #=> Integer
# resp.production_variants[0].instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.production_variants[0].initial_variant_weight #=> Float
# resp.production_variants[0].accelerator_type #=> String, one of "ml.eia1.medium", "ml.eia1.large", "ml.eia1.xlarge", "ml.eia2.medium", "ml.eia2.large", "ml.eia2.xlarge"
# resp.production_variants[0].core_dump_config.destination_s3_uri #=> String
# resp.production_variants[0].core_dump_config.kms_key_id #=> String
# resp.production_variants[0].serverless_config.memory_size_in_mb #=> Integer
# resp.production_variants[0].serverless_config.max_concurrency #=> Integer
# resp.production_variants[0].volume_size_in_gb #=> Integer
# resp.production_variants[0].model_data_download_timeout_in_seconds #=> Integer
# resp.production_variants[0].container_startup_health_check_timeout_in_seconds #=> Integer
# resp.production_variants[0].enable_ssm_access #=> Boolean
# resp.data_capture_config.enable_capture #=> Boolean
# resp.data_capture_config.initial_sampling_percentage #=> Integer
# resp.data_capture_config.destination_s3_uri #=> String
# resp.data_capture_config.kms_key_id #=> String
# resp.data_capture_config.capture_options #=> Array
# resp.data_capture_config.capture_options[0].capture_mode #=> String, one of "Input", "Output"
# resp.data_capture_config.capture_content_type_header.csv_content_types #=> Array
# resp.data_capture_config.capture_content_type_header.csv_content_types[0] #=> String
# resp.data_capture_config.capture_content_type_header.json_content_types #=> Array
# resp.data_capture_config.capture_content_type_header.json_content_types[0] #=> String
# resp.kms_key_id #=> String
# resp.creation_time #=> Time
# resp.async_inference_config.client_config.max_concurrent_invocations_per_instance #=> Integer
# resp.async_inference_config.output_config.kms_key_id #=> String
# resp.async_inference_config.output_config.s3_output_path #=> String
# resp.async_inference_config.output_config.notification_config.success_topic #=> String
# resp.async_inference_config.output_config.notification_config.error_topic #=> String
# resp.async_inference_config.output_config.notification_config.include_inference_response_in #=> Array
# resp.async_inference_config.output_config.notification_config.include_inference_response_in[0] #=> String, one of "SUCCESS_NOTIFICATION_TOPIC", "ERROR_NOTIFICATION_TOPIC"
# resp.async_inference_config.output_config.s3_failure_path #=> String
# resp.explainer_config.clarify_explainer_config.enable_explanations #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.features_attribute #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.content_template #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.max_record_count #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.max_payload_in_mb #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.probability_index #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.label_index #=> Integer
# resp.explainer_config.clarify_explainer_config.inference_config.probability_attribute #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.label_attribute #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.label_headers #=> Array
# resp.explainer_config.clarify_explainer_config.inference_config.label_headers[0] #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.feature_headers #=> Array
# resp.explainer_config.clarify_explainer_config.inference_config.feature_headers[0] #=> String
# resp.explainer_config.clarify_explainer_config.inference_config.feature_types #=> Array
# resp.explainer_config.clarify_explainer_config.inference_config.feature_types[0] #=> String, one of "numerical", "categorical", "text"
# resp.explainer_config.clarify_explainer_config.shap_config.shap_baseline_config.mime_type #=> String
# resp.explainer_config.clarify_explainer_config.shap_config.shap_baseline_config.shap_baseline #=> String
# resp.explainer_config.clarify_explainer_config.shap_config.shap_baseline_config.shap_baseline_uri #=> String
# resp.explainer_config.clarify_explainer_config.shap_config.number_of_samples #=> Integer
# resp.explainer_config.clarify_explainer_config.shap_config.use_logit #=> Boolean
# resp.explainer_config.clarify_explainer_config.shap_config.seed #=> Integer
# resp.explainer_config.clarify_explainer_config.shap_config.text_config.language #=> String, one of "af", "sq", "ar", "hy", "eu", "bn", "bg", "ca", "zh", "hr", "cs", "da", "nl", "en", "et", "fi", "fr", "de", "el", "gu", "he", "hi", "hu", "is", "id", "ga", "it", "kn", "ky", "lv", "lt", "lb", "mk", "ml", "mr", "ne", "nb", "fa", "pl", "pt", "ro", "ru", "sa", "sr", "tn", "si", "sk", "sl", "es", "sv", "tl", "ta", "tt", "te", "tr", "uk", "ur", "yo", "lij", "xx"
# resp.explainer_config.clarify_explainer_config.shap_config.text_config.granularity #=> String, one of "token", "sentence", "paragraph"
# resp.shadow_production_variants #=> Array
# resp.shadow_production_variants[0].variant_name #=> String
# resp.shadow_production_variants[0].model_name #=> String
# resp.shadow_production_variants[0].initial_instance_count #=> Integer
# resp.shadow_production_variants[0].instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.shadow_production_variants[0].initial_variant_weight #=> Float
# resp.shadow_production_variants[0].accelerator_type #=> String, one of "ml.eia1.medium", "ml.eia1.large", "ml.eia1.xlarge", "ml.eia2.medium", "ml.eia2.large", "ml.eia2.xlarge"
# resp.shadow_production_variants[0].core_dump_config.destination_s3_uri #=> String
# resp.shadow_production_variants[0].core_dump_config.kms_key_id #=> String
# resp.shadow_production_variants[0].serverless_config.memory_size_in_mb #=> Integer
# resp.shadow_production_variants[0].serverless_config.max_concurrency #=> Integer
# resp.shadow_production_variants[0].volume_size_in_gb #=> Integer
# resp.shadow_production_variants[0].model_data_download_timeout_in_seconds #=> Integer
# resp.shadow_production_variants[0].container_startup_health_check_timeout_in_seconds #=> Integer
# resp.shadow_production_variants[0].enable_ssm_access #=> Boolean
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeEndpointConfig AWS API Documentation
#
# @overload describe_endpoint_config(params = {})
# @param [Hash] params ({})
def describe_endpoint_config(params = {}, options = {})
req = build_request(:describe_endpoint_config, params)
req.send_request(options)
end
# Provides a list of an experiment's properties.
#
# @option params [required, String] :experiment_name
# The name of the experiment to describe.
#
# @return [Types::DescribeExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeExperimentResponse#experiment_name #experiment_name} => String
# * {Types::DescribeExperimentResponse#experiment_arn #experiment_arn} => String
# * {Types::DescribeExperimentResponse#display_name #display_name} => String
# * {Types::DescribeExperimentResponse#source #source} => Types::ExperimentSource
# * {Types::DescribeExperimentResponse#description #description} => String
# * {Types::DescribeExperimentResponse#creation_time #creation_time} => Time
# * {Types::DescribeExperimentResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeExperimentResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeExperimentResponse#last_modified_by #last_modified_by} => Types::UserContext
#
# @example Request syntax with placeholder values
#
# resp = client.describe_experiment({
# experiment_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.experiment_name #=> String
# resp.experiment_arn #=> String
# resp.display_name #=> String
# resp.source.source_arn #=> String
# resp.source.source_type #=> String
# resp.description #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeExperiment AWS API Documentation
#
# @overload describe_experiment(params = {})
# @param [Hash] params ({})
def describe_experiment(params = {}, options = {})
req = build_request(:describe_experiment, params)
req.send_request(options)
end
# Use this operation to describe a `FeatureGroup`. The response includes
# information on the creation time, `FeatureGroup` name, the unique
# identifier for each `FeatureGroup`, and more.
#
# @option params [required, String] :feature_group_name
# The name of the `FeatureGroup` you want described.
#
# @option params [String] :next_token
# A token to resume pagination of the list of `Features`
# (`FeatureDefinitions`). 2,500 `Features` are returned by default.
#
# @return [Types::DescribeFeatureGroupResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeFeatureGroupResponse#feature_group_arn #feature_group_arn} => String
# * {Types::DescribeFeatureGroupResponse#feature_group_name #feature_group_name} => String
# * {Types::DescribeFeatureGroupResponse#record_identifier_feature_name #record_identifier_feature_name} => String
# * {Types::DescribeFeatureGroupResponse#event_time_feature_name #event_time_feature_name} => String
# * {Types::DescribeFeatureGroupResponse#feature_definitions #feature_definitions} => Array<Types::FeatureDefinition>
# * {Types::DescribeFeatureGroupResponse#creation_time #creation_time} => Time
# * {Types::DescribeFeatureGroupResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeFeatureGroupResponse#online_store_config #online_store_config} => Types::OnlineStoreConfig
# * {Types::DescribeFeatureGroupResponse#offline_store_config #offline_store_config} => Types::OfflineStoreConfig
# * {Types::DescribeFeatureGroupResponse#role_arn #role_arn} => String
# * {Types::DescribeFeatureGroupResponse#feature_group_status #feature_group_status} => String
# * {Types::DescribeFeatureGroupResponse#offline_store_status #offline_store_status} => Types::OfflineStoreStatus
# * {Types::DescribeFeatureGroupResponse#last_update_status #last_update_status} => Types::LastUpdateStatus
# * {Types::DescribeFeatureGroupResponse#failure_reason #failure_reason} => String
# * {Types::DescribeFeatureGroupResponse#description #description} => String
# * {Types::DescribeFeatureGroupResponse#next_token #next_token} => String
# * {Types::DescribeFeatureGroupResponse#online_store_total_size_bytes #online_store_total_size_bytes} => Integer
#
# @example Request syntax with placeholder values
#
# resp = client.describe_feature_group({
# feature_group_name: "FeatureGroupName", # required
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.feature_group_arn #=> String
# resp.feature_group_name #=> String
# resp.record_identifier_feature_name #=> String
# resp.event_time_feature_name #=> String
# resp.feature_definitions #=> Array
# resp.feature_definitions[0].feature_name #=> String
# resp.feature_definitions[0].feature_type #=> String, one of "Integral", "Fractional", "String"
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.online_store_config.security_config.kms_key_id #=> String
# resp.online_store_config.enable_online_store #=> Boolean
# resp.offline_store_config.s3_storage_config.s3_uri #=> String
# resp.offline_store_config.s3_storage_config.kms_key_id #=> String
# resp.offline_store_config.s3_storage_config.resolved_output_s3_uri #=> String
# resp.offline_store_config.disable_glue_table_creation #=> Boolean
# resp.offline_store_config.data_catalog_config.table_name #=> String
# resp.offline_store_config.data_catalog_config.catalog #=> String
# resp.offline_store_config.data_catalog_config.database #=> String
# resp.offline_store_config.table_format #=> String, one of "Glue", "Iceberg"
# resp.role_arn #=> String
# resp.feature_group_status #=> String, one of "Creating", "Created", "CreateFailed", "Deleting", "DeleteFailed"
# resp.offline_store_status.status #=> String, one of "Active", "Blocked", "Disabled"
# resp.offline_store_status.blocked_reason #=> String
# resp.last_update_status.status #=> String, one of "Successful", "Failed", "InProgress"
# resp.last_update_status.failure_reason #=> String
# resp.failure_reason #=> String
# resp.description #=> String
# resp.next_token #=> String
# resp.online_store_total_size_bytes #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeFeatureGroup AWS API Documentation
#
# @overload describe_feature_group(params = {})
# @param [Hash] params ({})
def describe_feature_group(params = {}, options = {})
req = build_request(:describe_feature_group, params)
req.send_request(options)
end
# Shows the metadata for a feature within a feature group.
#
# @option params [required, String] :feature_group_name
# The name of the feature group containing the feature.
#
# @option params [required, String] :feature_name
# The name of the feature.
#
# @return [Types::DescribeFeatureMetadataResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeFeatureMetadataResponse#feature_group_arn #feature_group_arn} => String
# * {Types::DescribeFeatureMetadataResponse#feature_group_name #feature_group_name} => String
# * {Types::DescribeFeatureMetadataResponse#feature_name #feature_name} => String
# * {Types::DescribeFeatureMetadataResponse#feature_type #feature_type} => String
# * {Types::DescribeFeatureMetadataResponse#creation_time #creation_time} => Time
# * {Types::DescribeFeatureMetadataResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeFeatureMetadataResponse#description #description} => String
# * {Types::DescribeFeatureMetadataResponse#parameters #parameters} => Array<Types::FeatureParameter>
#
# @example Request syntax with placeholder values
#
# resp = client.describe_feature_metadata({
# feature_group_name: "FeatureGroupName", # required
# feature_name: "FeatureName", # required
# })
#
# @example Response structure
#
# resp.feature_group_arn #=> String
# resp.feature_group_name #=> String
# resp.feature_name #=> String
# resp.feature_type #=> String, one of "Integral", "Fractional", "String"
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.description #=> String
# resp.parameters #=> Array
# resp.parameters[0].key #=> String
# resp.parameters[0].value #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeFeatureMetadata AWS API Documentation
#
# @overload describe_feature_metadata(params = {})
# @param [Hash] params ({})
def describe_feature_metadata(params = {}, options = {})
req = build_request(:describe_feature_metadata, params)
req.send_request(options)
end
# Returns information about the specified flow definition.
#
# @option params [required, String] :flow_definition_name
# The name of the flow definition.
#
# @return [Types::DescribeFlowDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeFlowDefinitionResponse#flow_definition_arn #flow_definition_arn} => String
# * {Types::DescribeFlowDefinitionResponse#flow_definition_name #flow_definition_name} => String
# * {Types::DescribeFlowDefinitionResponse#flow_definition_status #flow_definition_status} => String
# * {Types::DescribeFlowDefinitionResponse#creation_time #creation_time} => Time
# * {Types::DescribeFlowDefinitionResponse#human_loop_request_source #human_loop_request_source} => Types::HumanLoopRequestSource
# * {Types::DescribeFlowDefinitionResponse#human_loop_activation_config #human_loop_activation_config} => Types::HumanLoopActivationConfig
# * {Types::DescribeFlowDefinitionResponse#human_loop_config #human_loop_config} => Types::HumanLoopConfig
# * {Types::DescribeFlowDefinitionResponse#output_config #output_config} => Types::FlowDefinitionOutputConfig
# * {Types::DescribeFlowDefinitionResponse#role_arn #role_arn} => String
# * {Types::DescribeFlowDefinitionResponse#failure_reason #failure_reason} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_flow_definition({
# flow_definition_name: "FlowDefinitionName", # required
# })
#
# @example Response structure
#
# resp.flow_definition_arn #=> String
# resp.flow_definition_name #=> String
# resp.flow_definition_status #=> String, one of "Initializing", "Active", "Failed", "Deleting"
# resp.creation_time #=> Time
# resp.human_loop_request_source.aws_managed_human_loop_request_source #=> String, one of "AWS/Rekognition/DetectModerationLabels/Image/V3", "AWS/Textract/AnalyzeDocument/Forms/V1"
# resp.human_loop_activation_config.human_loop_activation_conditions_config.human_loop_activation_conditions #=> String
# resp.human_loop_config.workteam_arn #=> String
# resp.human_loop_config.human_task_ui_arn #=> String
# resp.human_loop_config.task_title #=> String
# resp.human_loop_config.task_description #=> String
# resp.human_loop_config.task_count #=> Integer
# resp.human_loop_config.task_availability_lifetime_in_seconds #=> Integer
# resp.human_loop_config.task_time_limit_in_seconds #=> Integer
# resp.human_loop_config.task_keywords #=> Array
# resp.human_loop_config.task_keywords[0] #=> String
# resp.human_loop_config.public_workforce_task_price.amount_in_usd.dollars #=> Integer
# resp.human_loop_config.public_workforce_task_price.amount_in_usd.cents #=> Integer
# resp.human_loop_config.public_workforce_task_price.amount_in_usd.tenth_fractions_of_a_cent #=> Integer
# resp.output_config.s3_output_path #=> String
# resp.output_config.kms_key_id #=> String
# resp.role_arn #=> String
# resp.failure_reason #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeFlowDefinition AWS API Documentation
#
# @overload describe_flow_definition(params = {})
# @param [Hash] params ({})
def describe_flow_definition(params = {}, options = {})
req = build_request(:describe_flow_definition, params)
req.send_request(options)
end
# Describe a hub.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub to describe.
#
# @return [Types::DescribeHubResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeHubResponse#hub_name #hub_name} => String
# * {Types::DescribeHubResponse#hub_arn #hub_arn} => String
# * {Types::DescribeHubResponse#hub_display_name #hub_display_name} => String
# * {Types::DescribeHubResponse#hub_description #hub_description} => String
# * {Types::DescribeHubResponse#hub_search_keywords #hub_search_keywords} => Array<String>
# * {Types::DescribeHubResponse#s3_storage_config #s3_storage_config} => Types::HubS3StorageConfig
# * {Types::DescribeHubResponse#hub_status #hub_status} => String
# * {Types::DescribeHubResponse#failure_reason #failure_reason} => String
# * {Types::DescribeHubResponse#creation_time #creation_time} => Time
# * {Types::DescribeHubResponse#last_modified_time #last_modified_time} => Time
#
# @example Request syntax with placeholder values
#
# resp = client.describe_hub({
# hub_name: "HubName", # required
# })
#
# @example Response structure
#
# resp.hub_name #=> String
# resp.hub_arn #=> String
# resp.hub_display_name #=> String
# resp.hub_description #=> String
# resp.hub_search_keywords #=> Array
# resp.hub_search_keywords[0] #=> String
# resp.s3_storage_config.s3_output_path #=> String
# resp.hub_status #=> String, one of "InService", "Creating", "Updating", "Deleting", "CreateFailed", "UpdateFailed", "DeleteFailed"
# resp.failure_reason #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeHub AWS API Documentation
#
# @overload describe_hub(params = {})
# @param [Hash] params ({})
def describe_hub(params = {}, options = {})
req = build_request(:describe_hub, params)
req.send_request(options)
end
# Describe the content of a hub.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub that contains the content to describe.
#
# @option params [required, String] :hub_content_type
# The type of content in the hub.
#
# @option params [required, String] :hub_content_name
# The name of the content to describe.
#
# @option params [String] :hub_content_version
# The version of the content to describe.
#
# @return [Types::DescribeHubContentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeHubContentResponse#hub_content_name #hub_content_name} => String
# * {Types::DescribeHubContentResponse#hub_content_arn #hub_content_arn} => String
# * {Types::DescribeHubContentResponse#hub_content_version #hub_content_version} => String
# * {Types::DescribeHubContentResponse#hub_content_type #hub_content_type} => String
# * {Types::DescribeHubContentResponse#document_schema_version #document_schema_version} => String
# * {Types::DescribeHubContentResponse#hub_name #hub_name} => String
# * {Types::DescribeHubContentResponse#hub_arn #hub_arn} => String
# * {Types::DescribeHubContentResponse#hub_content_display_name #hub_content_display_name} => String
# * {Types::DescribeHubContentResponse#hub_content_description #hub_content_description} => String
# * {Types::DescribeHubContentResponse#hub_content_markdown #hub_content_markdown} => String
# * {Types::DescribeHubContentResponse#hub_content_document #hub_content_document} => String
# * {Types::DescribeHubContentResponse#hub_content_search_keywords #hub_content_search_keywords} => Array<String>
# * {Types::DescribeHubContentResponse#hub_content_dependencies #hub_content_dependencies} => Array<Types::HubContentDependency>
# * {Types::DescribeHubContentResponse#hub_content_status #hub_content_status} => String
# * {Types::DescribeHubContentResponse#failure_reason #failure_reason} => String
# * {Types::DescribeHubContentResponse#creation_time #creation_time} => Time
#
# @example Request syntax with placeholder values
#
# resp = client.describe_hub_content({
# hub_name: "HubName", # required
# hub_content_type: "Model", # required, accepts Model, Notebook
# hub_content_name: "HubContentName", # required
# hub_content_version: "HubContentVersion",
# })
#
# @example Response structure
#
# resp.hub_content_name #=> String
# resp.hub_content_arn #=> String
# resp.hub_content_version #=> String
# resp.hub_content_type #=> String, one of "Model", "Notebook"
# resp.document_schema_version #=> String
# resp.hub_name #=> String
# resp.hub_arn #=> String
# resp.hub_content_display_name #=> String
# resp.hub_content_description #=> String
# resp.hub_content_markdown #=> String
# resp.hub_content_document #=> String
# resp.hub_content_search_keywords #=> Array
# resp.hub_content_search_keywords[0] #=> String
# resp.hub_content_dependencies #=> Array
# resp.hub_content_dependencies[0].dependency_origin_path #=> String
# resp.hub_content_dependencies[0].dependency_copy_path #=> String
# resp.hub_content_status #=> String, one of "Available", "Importing", "Deleting", "ImportFailed", "DeleteFailed"
# resp.failure_reason #=> String
# resp.creation_time #=> Time
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeHubContent AWS API Documentation
#
# @overload describe_hub_content(params = {})
# @param [Hash] params ({})
def describe_hub_content(params = {}, options = {})
req = build_request(:describe_hub_content, params)
req.send_request(options)
end
# Returns information about the requested human task user interface
# (worker task template).
#
# @option params [required, String] :human_task_ui_name
# The name of the human task user interface (worker task template) you
# want information about.
#
# @return [Types::DescribeHumanTaskUiResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeHumanTaskUiResponse#human_task_ui_arn #human_task_ui_arn} => String
# * {Types::DescribeHumanTaskUiResponse#human_task_ui_name #human_task_ui_name} => String
# * {Types::DescribeHumanTaskUiResponse#human_task_ui_status #human_task_ui_status} => String
# * {Types::DescribeHumanTaskUiResponse#creation_time #creation_time} => Time
# * {Types::DescribeHumanTaskUiResponse#ui_template #ui_template} => Types::UiTemplateInfo
#
# @example Request syntax with placeholder values
#
# resp = client.describe_human_task_ui({
# human_task_ui_name: "HumanTaskUiName", # required
# })
#
# @example Response structure
#
# resp.human_task_ui_arn #=> String
# resp.human_task_ui_name #=> String
# resp.human_task_ui_status #=> String, one of "Active", "Deleting"
# resp.creation_time #=> Time
# resp.ui_template.url #=> String
# resp.ui_template.content_sha_256 #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeHumanTaskUi AWS API Documentation
#
# @overload describe_human_task_ui(params = {})
# @param [Hash] params ({})
def describe_human_task_ui(params = {}, options = {})
req = build_request(:describe_human_task_ui, params)
req.send_request(options)
end
# Gets a description of a hyperparameter tuning job.
#
# @option params [required, String] :hyper_parameter_tuning_job_name
# The name of the tuning job.
#
# @return [Types::DescribeHyperParameterTuningJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeHyperParameterTuningJobResponse#hyper_parameter_tuning_job_name #hyper_parameter_tuning_job_name} => String
# * {Types::DescribeHyperParameterTuningJobResponse#hyper_parameter_tuning_job_arn #hyper_parameter_tuning_job_arn} => String
# * {Types::DescribeHyperParameterTuningJobResponse#hyper_parameter_tuning_job_config #hyper_parameter_tuning_job_config} => Types::HyperParameterTuningJobConfig
# * {Types::DescribeHyperParameterTuningJobResponse#training_job_definition #training_job_definition} => Types::HyperParameterTrainingJobDefinition
# * {Types::DescribeHyperParameterTuningJobResponse#training_job_definitions #training_job_definitions} => Array<Types::HyperParameterTrainingJobDefinition>
# * {Types::DescribeHyperParameterTuningJobResponse#hyper_parameter_tuning_job_status #hyper_parameter_tuning_job_status} => String
# * {Types::DescribeHyperParameterTuningJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeHyperParameterTuningJobResponse#hyper_parameter_tuning_end_time #hyper_parameter_tuning_end_time} => Time
# * {Types::DescribeHyperParameterTuningJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeHyperParameterTuningJobResponse#training_job_status_counters #training_job_status_counters} => Types::TrainingJobStatusCounters
# * {Types::DescribeHyperParameterTuningJobResponse#objective_status_counters #objective_status_counters} => Types::ObjectiveStatusCounters
# * {Types::DescribeHyperParameterTuningJobResponse#best_training_job #best_training_job} => Types::HyperParameterTrainingJobSummary
# * {Types::DescribeHyperParameterTuningJobResponse#overall_best_training_job #overall_best_training_job} => Types::HyperParameterTrainingJobSummary
# * {Types::DescribeHyperParameterTuningJobResponse#warm_start_config #warm_start_config} => Types::HyperParameterTuningJobWarmStartConfig
# * {Types::DescribeHyperParameterTuningJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeHyperParameterTuningJobResponse#tuning_job_completion_details #tuning_job_completion_details} => Types::HyperParameterTuningJobCompletionDetails
# * {Types::DescribeHyperParameterTuningJobResponse#consumed_resources #consumed_resources} => Types::HyperParameterTuningJobConsumedResources
#
# @example Request syntax with placeholder values
#
# resp = client.describe_hyper_parameter_tuning_job({
# hyper_parameter_tuning_job_name: "HyperParameterTuningJobName", # required
# })
#
# @example Response structure
#
# resp.hyper_parameter_tuning_job_name #=> String
# resp.hyper_parameter_tuning_job_arn #=> String
# resp.hyper_parameter_tuning_job_config.strategy #=> String, one of "Bayesian", "Random", "Hyperband", "Grid"
# resp.hyper_parameter_tuning_job_config.strategy_config.hyperband_strategy_config.min_resource #=> Integer
# resp.hyper_parameter_tuning_job_config.strategy_config.hyperband_strategy_config.max_resource #=> Integer
# resp.hyper_parameter_tuning_job_config.hyper_parameter_tuning_job_objective.type #=> String, one of "Maximize", "Minimize"
# resp.hyper_parameter_tuning_job_config.hyper_parameter_tuning_job_objective.metric_name #=> String
# resp.hyper_parameter_tuning_job_config.resource_limits.max_number_of_training_jobs #=> Integer
# resp.hyper_parameter_tuning_job_config.resource_limits.max_parallel_training_jobs #=> Integer
# resp.hyper_parameter_tuning_job_config.resource_limits.max_runtime_in_seconds #=> Integer
# resp.hyper_parameter_tuning_job_config.parameter_ranges.integer_parameter_ranges #=> Array
# resp.hyper_parameter_tuning_job_config.parameter_ranges.integer_parameter_ranges[0].name #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.integer_parameter_ranges[0].min_value #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.integer_parameter_ranges[0].max_value #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.integer_parameter_ranges[0].scaling_type #=> String, one of "Auto", "Linear", "Logarithmic", "ReverseLogarithmic"
# resp.hyper_parameter_tuning_job_config.parameter_ranges.continuous_parameter_ranges #=> Array
# resp.hyper_parameter_tuning_job_config.parameter_ranges.continuous_parameter_ranges[0].name #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.continuous_parameter_ranges[0].min_value #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.continuous_parameter_ranges[0].max_value #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.continuous_parameter_ranges[0].scaling_type #=> String, one of "Auto", "Linear", "Logarithmic", "ReverseLogarithmic"
# resp.hyper_parameter_tuning_job_config.parameter_ranges.categorical_parameter_ranges #=> Array
# resp.hyper_parameter_tuning_job_config.parameter_ranges.categorical_parameter_ranges[0].name #=> String
# resp.hyper_parameter_tuning_job_config.parameter_ranges.categorical_parameter_ranges[0].values #=> Array
# resp.hyper_parameter_tuning_job_config.parameter_ranges.categorical_parameter_ranges[0].values[0] #=> String
# resp.hyper_parameter_tuning_job_config.training_job_early_stopping_type #=> String, one of "Off", "Auto"
# resp.hyper_parameter_tuning_job_config.tuning_job_completion_criteria.target_objective_metric_value #=> Float
# resp.hyper_parameter_tuning_job_config.tuning_job_completion_criteria.best_objective_not_improving.max_number_of_training_jobs_not_improving #=> Integer
# resp.hyper_parameter_tuning_job_config.tuning_job_completion_criteria.convergence_detected.complete_on_convergence #=> String, one of "Disabled", "Enabled"
# resp.hyper_parameter_tuning_job_config.random_seed #=> Integer
# resp.training_job_definition.definition_name #=> String
# resp.training_job_definition.tuning_objective.type #=> String, one of "Maximize", "Minimize"
# resp.training_job_definition.tuning_objective.metric_name #=> String
# resp.training_job_definition.hyper_parameter_ranges.integer_parameter_ranges #=> Array
# resp.training_job_definition.hyper_parameter_ranges.integer_parameter_ranges[0].name #=> String
# resp.training_job_definition.hyper_parameter_ranges.integer_parameter_ranges[0].min_value #=> String
# resp.training_job_definition.hyper_parameter_ranges.integer_parameter_ranges[0].max_value #=> String
# resp.training_job_definition.hyper_parameter_ranges.integer_parameter_ranges[0].scaling_type #=> String, one of "Auto", "Linear", "Logarithmic", "ReverseLogarithmic"
# resp.training_job_definition.hyper_parameter_ranges.continuous_parameter_ranges #=> Array
# resp.training_job_definition.hyper_parameter_ranges.continuous_parameter_ranges[0].name #=> String
# resp.training_job_definition.hyper_parameter_ranges.continuous_parameter_ranges[0].min_value #=> String
# resp.training_job_definition.hyper_parameter_ranges.continuous_parameter_ranges[0].max_value #=> String
# resp.training_job_definition.hyper_parameter_ranges.continuous_parameter_ranges[0].scaling_type #=> String, one of "Auto", "Linear", "Logarithmic", "ReverseLogarithmic"
# resp.training_job_definition.hyper_parameter_ranges.categorical_parameter_ranges #=> Array
# resp.training_job_definition.hyper_parameter_ranges.categorical_parameter_ranges[0].name #=> String
# resp.training_job_definition.hyper_parameter_ranges.categorical_parameter_ranges[0].values #=> Array
# resp.training_job_definition.hyper_parameter_ranges.categorical_parameter_ranges[0].values[0] #=> String
# resp.training_job_definition.static_hyper_parameters #=> Hash
# resp.training_job_definition.static_hyper_parameters["HyperParameterKey"] #=> String
# resp.training_job_definition.algorithm_specification.training_image #=> String
# resp.training_job_definition.algorithm_specification.training_input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.training_job_definition.algorithm_specification.algorithm_name #=> String
# resp.training_job_definition.algorithm_specification.metric_definitions #=> Array
# resp.training_job_definition.algorithm_specification.metric_definitions[0].name #=> String
# resp.training_job_definition.algorithm_specification.metric_definitions[0].regex #=> String
# resp.training_job_definition.role_arn #=> String
# resp.training_job_definition.input_data_config #=> Array
# resp.training_job_definition.input_data_config[0].channel_name #=> String
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.s3_uri #=> String
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.attribute_names #=> Array
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.attribute_names[0] #=> String
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.instance_group_names #=> Array
# resp.training_job_definition.input_data_config[0].data_source.s3_data_source.instance_group_names[0] #=> String
# resp.training_job_definition.input_data_config[0].data_source.file_system_data_source.file_system_id #=> String
# resp.training_job_definition.input_data_config[0].data_source.file_system_data_source.file_system_access_mode #=> String, one of "rw", "ro"
# resp.training_job_definition.input_data_config[0].data_source.file_system_data_source.file_system_type #=> String, one of "EFS", "FSxLustre"
# resp.training_job_definition.input_data_config[0].data_source.file_system_data_source.directory_path #=> String
# resp.training_job_definition.input_data_config[0].content_type #=> String
# resp.training_job_definition.input_data_config[0].compression_type #=> String, one of "None", "Gzip"
# resp.training_job_definition.input_data_config[0].record_wrapper_type #=> String, one of "None", "RecordIO"
# resp.training_job_definition.input_data_config[0].input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.training_job_definition.input_data_config[0].shuffle_config.seed #=> Integer
# resp.training_job_definition.vpc_config.security_group_ids #=> Array
# resp.training_job_definition.vpc_config.security_group_ids[0] #=> String
# resp.training_job_definition.vpc_config.subnets #=> Array
# resp.training_job_definition.vpc_config.subnets[0] #=> String
# resp.training_job_definition.output_data_config.kms_key_id #=> String
# resp.training_job_definition.output_data_config.s3_output_path #=> String
# resp.training_job_definition.resource_config.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definition.resource_config.instance_count #=> Integer
# resp.training_job_definition.resource_config.volume_size_in_gb #=> Integer
# resp.training_job_definition.resource_config.volume_kms_key_id #=> String
# resp.training_job_definition.resource_config.instance_groups #=> Array
# resp.training_job_definition.resource_config.instance_groups[0].instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definition.resource_config.instance_groups[0].instance_count #=> Integer
# resp.training_job_definition.resource_config.instance_groups[0].instance_group_name #=> String
# resp.training_job_definition.resource_config.keep_alive_period_in_seconds #=> Integer
# resp.training_job_definition.stopping_condition.max_runtime_in_seconds #=> Integer
# resp.training_job_definition.stopping_condition.max_wait_time_in_seconds #=> Integer
# resp.training_job_definition.enable_network_isolation #=> Boolean
# resp.training_job_definition.enable_inter_container_traffic_encryption #=> Boolean
# resp.training_job_definition.enable_managed_spot_training #=> Boolean
# resp.training_job_definition.checkpoint_config.s3_uri #=> String
# resp.training_job_definition.checkpoint_config.local_path #=> String
# resp.training_job_definition.retry_strategy.maximum_retry_attempts #=> Integer
# resp.training_job_definition.hyper_parameter_tuning_resource_config.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definition.hyper_parameter_tuning_resource_config.instance_count #=> Integer
# resp.training_job_definition.hyper_parameter_tuning_resource_config.volume_size_in_gb #=> Integer
# resp.training_job_definition.hyper_parameter_tuning_resource_config.volume_kms_key_id #=> String
# resp.training_job_definition.hyper_parameter_tuning_resource_config.allocation_strategy #=> String, one of "Prioritized"
# resp.training_job_definition.hyper_parameter_tuning_resource_config.instance_configs #=> Array
# resp.training_job_definition.hyper_parameter_tuning_resource_config.instance_configs[0].instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definition.hyper_parameter_tuning_resource_config.instance_configs[0].instance_count #=> Integer
# resp.training_job_definition.hyper_parameter_tuning_resource_config.instance_configs[0].volume_size_in_gb #=> Integer
# resp.training_job_definition.environment #=> Hash
# resp.training_job_definition.environment["HyperParameterTrainingJobEnvironmentKey"] #=> String
# resp.training_job_definitions #=> Array
# resp.training_job_definitions[0].definition_name #=> String
# resp.training_job_definitions[0].tuning_objective.type #=> String, one of "Maximize", "Minimize"
# resp.training_job_definitions[0].tuning_objective.metric_name #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.integer_parameter_ranges #=> Array
# resp.training_job_definitions[0].hyper_parameter_ranges.integer_parameter_ranges[0].name #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.integer_parameter_ranges[0].min_value #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.integer_parameter_ranges[0].max_value #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.integer_parameter_ranges[0].scaling_type #=> String, one of "Auto", "Linear", "Logarithmic", "ReverseLogarithmic"
# resp.training_job_definitions[0].hyper_parameter_ranges.continuous_parameter_ranges #=> Array
# resp.training_job_definitions[0].hyper_parameter_ranges.continuous_parameter_ranges[0].name #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.continuous_parameter_ranges[0].min_value #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.continuous_parameter_ranges[0].max_value #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.continuous_parameter_ranges[0].scaling_type #=> String, one of "Auto", "Linear", "Logarithmic", "ReverseLogarithmic"
# resp.training_job_definitions[0].hyper_parameter_ranges.categorical_parameter_ranges #=> Array
# resp.training_job_definitions[0].hyper_parameter_ranges.categorical_parameter_ranges[0].name #=> String
# resp.training_job_definitions[0].hyper_parameter_ranges.categorical_parameter_ranges[0].values #=> Array
# resp.training_job_definitions[0].hyper_parameter_ranges.categorical_parameter_ranges[0].values[0] #=> String
# resp.training_job_definitions[0].static_hyper_parameters #=> Hash
# resp.training_job_definitions[0].static_hyper_parameters["HyperParameterKey"] #=> String
# resp.training_job_definitions[0].algorithm_specification.training_image #=> String
# resp.training_job_definitions[0].algorithm_specification.training_input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.training_job_definitions[0].algorithm_specification.algorithm_name #=> String
# resp.training_job_definitions[0].algorithm_specification.metric_definitions #=> Array
# resp.training_job_definitions[0].algorithm_specification.metric_definitions[0].name #=> String
# resp.training_job_definitions[0].algorithm_specification.metric_definitions[0].regex #=> String
# resp.training_job_definitions[0].role_arn #=> String
# resp.training_job_definitions[0].input_data_config #=> Array
# resp.training_job_definitions[0].input_data_config[0].channel_name #=> String
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.s3_uri #=> String
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.attribute_names #=> Array
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.attribute_names[0] #=> String
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.instance_group_names #=> Array
# resp.training_job_definitions[0].input_data_config[0].data_source.s3_data_source.instance_group_names[0] #=> String
# resp.training_job_definitions[0].input_data_config[0].data_source.file_system_data_source.file_system_id #=> String
# resp.training_job_definitions[0].input_data_config[0].data_source.file_system_data_source.file_system_access_mode #=> String, one of "rw", "ro"
# resp.training_job_definitions[0].input_data_config[0].data_source.file_system_data_source.file_system_type #=> String, one of "EFS", "FSxLustre"
# resp.training_job_definitions[0].input_data_config[0].data_source.file_system_data_source.directory_path #=> String
# resp.training_job_definitions[0].input_data_config[0].content_type #=> String
# resp.training_job_definitions[0].input_data_config[0].compression_type #=> String, one of "None", "Gzip"
# resp.training_job_definitions[0].input_data_config[0].record_wrapper_type #=> String, one of "None", "RecordIO"
# resp.training_job_definitions[0].input_data_config[0].input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.training_job_definitions[0].input_data_config[0].shuffle_config.seed #=> Integer
# resp.training_job_definitions[0].vpc_config.security_group_ids #=> Array
# resp.training_job_definitions[0].vpc_config.security_group_ids[0] #=> String
# resp.training_job_definitions[0].vpc_config.subnets #=> Array
# resp.training_job_definitions[0].vpc_config.subnets[0] #=> String
# resp.training_job_definitions[0].output_data_config.kms_key_id #=> String
# resp.training_job_definitions[0].output_data_config.s3_output_path #=> String
# resp.training_job_definitions[0].resource_config.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definitions[0].resource_config.instance_count #=> Integer
# resp.training_job_definitions[0].resource_config.volume_size_in_gb #=> Integer
# resp.training_job_definitions[0].resource_config.volume_kms_key_id #=> String
# resp.training_job_definitions[0].resource_config.instance_groups #=> Array
# resp.training_job_definitions[0].resource_config.instance_groups[0].instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definitions[0].resource_config.instance_groups[0].instance_count #=> Integer
# resp.training_job_definitions[0].resource_config.instance_groups[0].instance_group_name #=> String
# resp.training_job_definitions[0].resource_config.keep_alive_period_in_seconds #=> Integer
# resp.training_job_definitions[0].stopping_condition.max_runtime_in_seconds #=> Integer
# resp.training_job_definitions[0].stopping_condition.max_wait_time_in_seconds #=> Integer
# resp.training_job_definitions[0].enable_network_isolation #=> Boolean
# resp.training_job_definitions[0].enable_inter_container_traffic_encryption #=> Boolean
# resp.training_job_definitions[0].enable_managed_spot_training #=> Boolean
# resp.training_job_definitions[0].checkpoint_config.s3_uri #=> String
# resp.training_job_definitions[0].checkpoint_config.local_path #=> String
# resp.training_job_definitions[0].retry_strategy.maximum_retry_attempts #=> Integer
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.instance_count #=> Integer
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.volume_size_in_gb #=> Integer
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.volume_kms_key_id #=> String
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.allocation_strategy #=> String, one of "Prioritized"
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.instance_configs #=> Array
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.instance_configs[0].instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.instance_configs[0].instance_count #=> Integer
# resp.training_job_definitions[0].hyper_parameter_tuning_resource_config.instance_configs[0].volume_size_in_gb #=> Integer
# resp.training_job_definitions[0].environment #=> Hash
# resp.training_job_definitions[0].environment["HyperParameterTrainingJobEnvironmentKey"] #=> String
# resp.hyper_parameter_tuning_job_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.creation_time #=> Time
# resp.hyper_parameter_tuning_end_time #=> Time
# resp.last_modified_time #=> Time
# resp.training_job_status_counters.completed #=> Integer
# resp.training_job_status_counters.in_progress #=> Integer
# resp.training_job_status_counters.retryable_error #=> Integer
# resp.training_job_status_counters.non_retryable_error #=> Integer
# resp.training_job_status_counters.stopped #=> Integer
# resp.objective_status_counters.succeeded #=> Integer
# resp.objective_status_counters.pending #=> Integer
# resp.objective_status_counters.failed #=> Integer
# resp.best_training_job.training_job_definition_name #=> String
# resp.best_training_job.training_job_name #=> String
# resp.best_training_job.training_job_arn #=> String
# resp.best_training_job.tuning_job_name #=> String
# resp.best_training_job.creation_time #=> Time
# resp.best_training_job.training_start_time #=> Time
# resp.best_training_job.training_end_time #=> Time
# resp.best_training_job.training_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.best_training_job.tuned_hyper_parameters #=> Hash
# resp.best_training_job.tuned_hyper_parameters["HyperParameterKey"] #=> String
# resp.best_training_job.failure_reason #=> String
# resp.best_training_job.final_hyper_parameter_tuning_job_objective_metric.type #=> String, one of "Maximize", "Minimize"
# resp.best_training_job.final_hyper_parameter_tuning_job_objective_metric.metric_name #=> String
# resp.best_training_job.final_hyper_parameter_tuning_job_objective_metric.value #=> Float
# resp.best_training_job.objective_status #=> String, one of "Succeeded", "Pending", "Failed"
# resp.overall_best_training_job.training_job_definition_name #=> String
# resp.overall_best_training_job.training_job_name #=> String
# resp.overall_best_training_job.training_job_arn #=> String
# resp.overall_best_training_job.tuning_job_name #=> String
# resp.overall_best_training_job.creation_time #=> Time
# resp.overall_best_training_job.training_start_time #=> Time
# resp.overall_best_training_job.training_end_time #=> Time
# resp.overall_best_training_job.training_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.overall_best_training_job.tuned_hyper_parameters #=> Hash
# resp.overall_best_training_job.tuned_hyper_parameters["HyperParameterKey"] #=> String
# resp.overall_best_training_job.failure_reason #=> String
# resp.overall_best_training_job.final_hyper_parameter_tuning_job_objective_metric.type #=> String, one of "Maximize", "Minimize"
# resp.overall_best_training_job.final_hyper_parameter_tuning_job_objective_metric.metric_name #=> String
# resp.overall_best_training_job.final_hyper_parameter_tuning_job_objective_metric.value #=> Float
# resp.overall_best_training_job.objective_status #=> String, one of "Succeeded", "Pending", "Failed"
# resp.warm_start_config.parent_hyper_parameter_tuning_jobs #=> Array
# resp.warm_start_config.parent_hyper_parameter_tuning_jobs[0].hyper_parameter_tuning_job_name #=> String
# resp.warm_start_config.warm_start_type #=> String, one of "IdenticalDataAndAlgorithm", "TransferLearning"
# resp.failure_reason #=> String
# resp.tuning_job_completion_details.number_of_training_jobs_objective_not_improving #=> Integer
# resp.tuning_job_completion_details.convergence_detected_time #=> Time
# resp.consumed_resources.runtime_in_seconds #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeHyperParameterTuningJob AWS API Documentation
#
# @overload describe_hyper_parameter_tuning_job(params = {})
# @param [Hash] params ({})
def describe_hyper_parameter_tuning_job(params = {}, options = {})
req = build_request(:describe_hyper_parameter_tuning_job, params)
req.send_request(options)
end
# Describes a SageMaker image.
#
# @option params [required, String] :image_name
# The name of the image to describe.
#
# @return [Types::DescribeImageResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeImageResponse#creation_time #creation_time} => Time
# * {Types::DescribeImageResponse#description #description} => String
# * {Types::DescribeImageResponse#display_name #display_name} => String
# * {Types::DescribeImageResponse#failure_reason #failure_reason} => String
# * {Types::DescribeImageResponse#image_arn #image_arn} => String
# * {Types::DescribeImageResponse#image_name #image_name} => String
# * {Types::DescribeImageResponse#image_status #image_status} => String
# * {Types::DescribeImageResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeImageResponse#role_arn #role_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_image({
# image_name: "ImageName", # required
# })
#
# @example Response structure
#
# resp.creation_time #=> Time
# resp.description #=> String
# resp.display_name #=> String
# resp.failure_reason #=> String
# resp.image_arn #=> String
# resp.image_name #=> String
# resp.image_status #=> String, one of "CREATING", "CREATED", "CREATE_FAILED", "UPDATING", "UPDATE_FAILED", "DELETING", "DELETE_FAILED"
# resp.last_modified_time #=> Time
# resp.role_arn #=> String
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * image_created
# * image_deleted
# * image_updated
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeImage AWS API Documentation
#
# @overload describe_image(params = {})
# @param [Hash] params ({})
def describe_image(params = {}, options = {})
req = build_request(:describe_image, params)
req.send_request(options)
end
# Describes a version of a SageMaker image.
#
# @option params [required, String] :image_name
# The name of the image.
#
# @option params [Integer] :version
# The version of the image. If not specified, the latest version is
# described.
#
# @option params [String] :alias
# The alias of the image version.
#
# @return [Types::DescribeImageVersionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeImageVersionResponse#base_image #base_image} => String
# * {Types::DescribeImageVersionResponse#container_image #container_image} => String
# * {Types::DescribeImageVersionResponse#creation_time #creation_time} => Time
# * {Types::DescribeImageVersionResponse#failure_reason #failure_reason} => String
# * {Types::DescribeImageVersionResponse#image_arn #image_arn} => String
# * {Types::DescribeImageVersionResponse#image_version_arn #image_version_arn} => String
# * {Types::DescribeImageVersionResponse#image_version_status #image_version_status} => String
# * {Types::DescribeImageVersionResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeImageVersionResponse#version #version} => Integer
# * {Types::DescribeImageVersionResponse#vendor_guidance #vendor_guidance} => String
# * {Types::DescribeImageVersionResponse#job_type #job_type} => String
# * {Types::DescribeImageVersionResponse#ml_framework #ml_framework} => String
# * {Types::DescribeImageVersionResponse#programming_lang #programming_lang} => String
# * {Types::DescribeImageVersionResponse#processor #processor} => String
# * {Types::DescribeImageVersionResponse#horovod #horovod} => Boolean
# * {Types::DescribeImageVersionResponse#release_notes #release_notes} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_image_version({
# image_name: "ImageName", # required
# version: 1,
# alias: "SageMakerImageVersionAlias",
# })
#
# @example Response structure
#
# resp.base_image #=> String
# resp.container_image #=> String
# resp.creation_time #=> Time
# resp.failure_reason #=> String
# resp.image_arn #=> String
# resp.image_version_arn #=> String
# resp.image_version_status #=> String, one of "CREATING", "CREATED", "CREATE_FAILED", "DELETING", "DELETE_FAILED"
# resp.last_modified_time #=> Time
# resp.version #=> Integer
# resp.vendor_guidance #=> String, one of "NOT_PROVIDED", "STABLE", "TO_BE_ARCHIVED", "ARCHIVED"
# resp.job_type #=> String, one of "TRAINING", "INFERENCE", "NOTEBOOK_KERNEL"
# resp.ml_framework #=> String
# resp.programming_lang #=> String
# resp.processor #=> String, one of "CPU", "GPU"
# resp.horovod #=> Boolean
# resp.release_notes #=> String
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * image_version_created
# * image_version_deleted
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeImageVersion AWS API Documentation
#
# @overload describe_image_version(params = {})
# @param [Hash] params ({})
def describe_image_version(params = {}, options = {})
req = build_request(:describe_image_version, params)
req.send_request(options)
end
# Returns details about an inference experiment.
#
# @option params [required, String] :name
# The name of the inference experiment to describe.
#
# @return [Types::DescribeInferenceExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeInferenceExperimentResponse#arn #arn} => String
# * {Types::DescribeInferenceExperimentResponse#name #name} => String
# * {Types::DescribeInferenceExperimentResponse#type #type} => String
# * {Types::DescribeInferenceExperimentResponse#schedule #schedule} => Types::InferenceExperimentSchedule
# * {Types::DescribeInferenceExperimentResponse#status #status} => String
# * {Types::DescribeInferenceExperimentResponse#status_reason #status_reason} => String
# * {Types::DescribeInferenceExperimentResponse#description #description} => String
# * {Types::DescribeInferenceExperimentResponse#creation_time #creation_time} => Time
# * {Types::DescribeInferenceExperimentResponse#completion_time #completion_time} => Time
# * {Types::DescribeInferenceExperimentResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeInferenceExperimentResponse#role_arn #role_arn} => String
# * {Types::DescribeInferenceExperimentResponse#endpoint_metadata #endpoint_metadata} => Types::EndpointMetadata
# * {Types::DescribeInferenceExperimentResponse#model_variants #model_variants} => Array<Types::ModelVariantConfigSummary>
# * {Types::DescribeInferenceExperimentResponse#data_storage_config #data_storage_config} => Types::InferenceExperimentDataStorageConfig
# * {Types::DescribeInferenceExperimentResponse#shadow_mode_config #shadow_mode_config} => Types::ShadowModeConfig
# * {Types::DescribeInferenceExperimentResponse#kms_key #kms_key} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_inference_experiment({
# name: "InferenceExperimentName", # required
# })
#
# @example Response structure
#
# resp.arn #=> String
# resp.name #=> String
# resp.type #=> String, one of "ShadowMode"
# resp.schedule.start_time #=> Time
# resp.schedule.end_time #=> Time
# resp.status #=> String, one of "Creating", "Created", "Updating", "Running", "Starting", "Stopping", "Completed", "Cancelled"
# resp.status_reason #=> String
# resp.description #=> String
# resp.creation_time #=> Time
# resp.completion_time #=> Time
# resp.last_modified_time #=> Time
# resp.role_arn #=> String
# resp.endpoint_metadata.endpoint_name #=> String
# resp.endpoint_metadata.endpoint_config_name #=> String
# resp.endpoint_metadata.endpoint_status #=> String, one of "OutOfService", "Creating", "Updating", "SystemUpdating", "RollingBack", "InService", "Deleting", "Failed"
# resp.endpoint_metadata.failure_reason #=> String
# resp.model_variants #=> Array
# resp.model_variants[0].model_name #=> String
# resp.model_variants[0].variant_name #=> String
# resp.model_variants[0].infrastructure_config.infrastructure_type #=> String, one of "RealTimeInference"
# resp.model_variants[0].infrastructure_config.real_time_inference_config.instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge"
# resp.model_variants[0].infrastructure_config.real_time_inference_config.instance_count #=> Integer
# resp.model_variants[0].status #=> String, one of "Creating", "Updating", "InService", "Deleting", "Deleted"
# resp.data_storage_config.destination #=> String
# resp.data_storage_config.kms_key #=> String
# resp.data_storage_config.content_type.csv_content_types #=> Array
# resp.data_storage_config.content_type.csv_content_types[0] #=> String
# resp.data_storage_config.content_type.json_content_types #=> Array
# resp.data_storage_config.content_type.json_content_types[0] #=> String
# resp.shadow_mode_config.source_model_variant_name #=> String
# resp.shadow_mode_config.shadow_model_variants #=> Array
# resp.shadow_mode_config.shadow_model_variants[0].shadow_model_variant_name #=> String
# resp.shadow_mode_config.shadow_model_variants[0].sampling_percentage #=> Integer
# resp.kms_key #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeInferenceExperiment AWS API Documentation
#
# @overload describe_inference_experiment(params = {})
# @param [Hash] params ({})
def describe_inference_experiment(params = {}, options = {})
req = build_request(:describe_inference_experiment, params)
req.send_request(options)
end
# Provides the results of the Inference Recommender job. One or more
# recommendation jobs are returned.
#
# @option params [required, String] :job_name
# The name of the job. The name must be unique within an Amazon Web
# Services Region in the Amazon Web Services account.
#
# @return [Types::DescribeInferenceRecommendationsJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeInferenceRecommendationsJobResponse#job_name #job_name} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#job_description #job_description} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#job_type #job_type} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#job_arn #job_arn} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#role_arn #role_arn} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#status #status} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeInferenceRecommendationsJobResponse#completion_time #completion_time} => Time
# * {Types::DescribeInferenceRecommendationsJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeInferenceRecommendationsJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeInferenceRecommendationsJobResponse#input_config #input_config} => Types::RecommendationJobInputConfig
# * {Types::DescribeInferenceRecommendationsJobResponse#stopping_conditions #stopping_conditions} => Types::RecommendationJobStoppingConditions
# * {Types::DescribeInferenceRecommendationsJobResponse#inference_recommendations #inference_recommendations} => Array<Types::InferenceRecommendation>
# * {Types::DescribeInferenceRecommendationsJobResponse#endpoint_performances #endpoint_performances} => Array<Types::EndpointPerformance>
#
# @example Request syntax with placeholder values
#
# resp = client.describe_inference_recommendations_job({
# job_name: "RecommendationJobName", # required
# })
#
# @example Response structure
#
# resp.job_name #=> String
# resp.job_description #=> String
# resp.job_type #=> String, one of "Default", "Advanced"
# resp.job_arn #=> String
# resp.role_arn #=> String
# resp.status #=> String, one of "PENDING", "IN_PROGRESS", "COMPLETED", "FAILED", "STOPPING", "STOPPED"
# resp.creation_time #=> Time
# resp.completion_time #=> Time
# resp.last_modified_time #=> Time
# resp.failure_reason #=> String
# resp.input_config.model_package_version_arn #=> String
# resp.input_config.job_duration_in_seconds #=> Integer
# resp.input_config.traffic_pattern.traffic_type #=> String, one of "PHASES"
# resp.input_config.traffic_pattern.phases #=> Array
# resp.input_config.traffic_pattern.phases[0].initial_number_of_users #=> Integer
# resp.input_config.traffic_pattern.phases[0].spawn_rate #=> Integer
# resp.input_config.traffic_pattern.phases[0].duration_in_seconds #=> Integer
# resp.input_config.resource_limit.max_number_of_tests #=> Integer
# resp.input_config.resource_limit.max_parallel_of_tests #=> Integer
# resp.input_config.endpoint_configurations #=> Array
# resp.input_config.endpoint_configurations[0].instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.input_config.endpoint_configurations[0].inference_specification_name #=> String
# resp.input_config.endpoint_configurations[0].environment_parameter_ranges.categorical_parameter_ranges #=> Array
# resp.input_config.endpoint_configurations[0].environment_parameter_ranges.categorical_parameter_ranges[0].name #=> String
# resp.input_config.endpoint_configurations[0].environment_parameter_ranges.categorical_parameter_ranges[0].value #=> Array
# resp.input_config.endpoint_configurations[0].environment_parameter_ranges.categorical_parameter_ranges[0].value[0] #=> String
# resp.input_config.volume_kms_key_id #=> String
# resp.input_config.container_config.domain #=> String
# resp.input_config.container_config.task #=> String
# resp.input_config.container_config.framework #=> String
# resp.input_config.container_config.framework_version #=> String
# resp.input_config.container_config.payload_config.sample_payload_url #=> String
# resp.input_config.container_config.payload_config.supported_content_types #=> Array
# resp.input_config.container_config.payload_config.supported_content_types[0] #=> String
# resp.input_config.container_config.nearest_model_name #=> String
# resp.input_config.container_config.supported_instance_types #=> Array
# resp.input_config.container_config.supported_instance_types[0] #=> String
# resp.input_config.container_config.data_input_config #=> String
# resp.input_config.endpoints #=> Array
# resp.input_config.endpoints[0].endpoint_name #=> String
# resp.input_config.vpc_config.security_group_ids #=> Array
# resp.input_config.vpc_config.security_group_ids[0] #=> String
# resp.input_config.vpc_config.subnets #=> Array
# resp.input_config.vpc_config.subnets[0] #=> String
# resp.input_config.model_name #=> String
# resp.stopping_conditions.max_invocations #=> Integer
# resp.stopping_conditions.model_latency_thresholds #=> Array
# resp.stopping_conditions.model_latency_thresholds[0].percentile #=> String
# resp.stopping_conditions.model_latency_thresholds[0].value_in_milliseconds #=> Integer
# resp.inference_recommendations #=> Array
# resp.inference_recommendations[0].metrics.cost_per_hour #=> Float
# resp.inference_recommendations[0].metrics.cost_per_inference #=> Float
# resp.inference_recommendations[0].metrics.max_invocations #=> Integer
# resp.inference_recommendations[0].metrics.model_latency #=> Integer
# resp.inference_recommendations[0].metrics.cpu_utilization #=> Float
# resp.inference_recommendations[0].metrics.memory_utilization #=> Float
# resp.inference_recommendations[0].endpoint_configuration.endpoint_name #=> String
# resp.inference_recommendations[0].endpoint_configuration.variant_name #=> String
# resp.inference_recommendations[0].endpoint_configuration.instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.inference_recommendations[0].endpoint_configuration.initial_instance_count #=> Integer
# resp.inference_recommendations[0].model_configuration.inference_specification_name #=> String
# resp.inference_recommendations[0].model_configuration.environment_parameters #=> Array
# resp.inference_recommendations[0].model_configuration.environment_parameters[0].key #=> String
# resp.inference_recommendations[0].model_configuration.environment_parameters[0].value_type #=> String
# resp.inference_recommendations[0].model_configuration.environment_parameters[0].value #=> String
# resp.inference_recommendations[0].model_configuration.compilation_job_name #=> String
# resp.inference_recommendations[0].recommendation_id #=> String
# resp.endpoint_performances #=> Array
# resp.endpoint_performances[0].metrics.max_invocations #=> Integer
# resp.endpoint_performances[0].metrics.model_latency #=> Integer
# resp.endpoint_performances[0].endpoint_info.endpoint_name #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeInferenceRecommendationsJob AWS API Documentation
#
# @overload describe_inference_recommendations_job(params = {})
# @param [Hash] params ({})
def describe_inference_recommendations_job(params = {}, options = {})
req = build_request(:describe_inference_recommendations_job, params)
req.send_request(options)
end
# Gets information about a labeling job.
#
# @option params [required, String] :labeling_job_name
# The name of the labeling job to return information for.
#
# @return [Types::DescribeLabelingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeLabelingJobResponse#labeling_job_status #labeling_job_status} => String
# * {Types::DescribeLabelingJobResponse#label_counters #label_counters} => Types::LabelCounters
# * {Types::DescribeLabelingJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeLabelingJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeLabelingJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeLabelingJobResponse#job_reference_code #job_reference_code} => String
# * {Types::DescribeLabelingJobResponse#labeling_job_name #labeling_job_name} => String
# * {Types::DescribeLabelingJobResponse#labeling_job_arn #labeling_job_arn} => String
# * {Types::DescribeLabelingJobResponse#label_attribute_name #label_attribute_name} => String
# * {Types::DescribeLabelingJobResponse#input_config #input_config} => Types::LabelingJobInputConfig
# * {Types::DescribeLabelingJobResponse#output_config #output_config} => Types::LabelingJobOutputConfig
# * {Types::DescribeLabelingJobResponse#role_arn #role_arn} => String
# * {Types::DescribeLabelingJobResponse#label_category_config_s3_uri #label_category_config_s3_uri} => String
# * {Types::DescribeLabelingJobResponse#stopping_conditions #stopping_conditions} => Types::LabelingJobStoppingConditions
# * {Types::DescribeLabelingJobResponse#labeling_job_algorithms_config #labeling_job_algorithms_config} => Types::LabelingJobAlgorithmsConfig
# * {Types::DescribeLabelingJobResponse#human_task_config #human_task_config} => Types::HumanTaskConfig
# * {Types::DescribeLabelingJobResponse#tags #tags} => Array<Types::Tag>
# * {Types::DescribeLabelingJobResponse#labeling_job_output #labeling_job_output} => Types::LabelingJobOutput
#
# @example Request syntax with placeholder values
#
# resp = client.describe_labeling_job({
# labeling_job_name: "LabelingJobName", # required
# })
#
# @example Response structure
#
# resp.labeling_job_status #=> String, one of "Initializing", "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.label_counters.total_labeled #=> Integer
# resp.label_counters.human_labeled #=> Integer
# resp.label_counters.machine_labeled #=> Integer
# resp.label_counters.failed_non_retryable_error #=> Integer
# resp.label_counters.unlabeled #=> Integer
# resp.failure_reason #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.job_reference_code #=> String
# resp.labeling_job_name #=> String
# resp.labeling_job_arn #=> String
# resp.label_attribute_name #=> String
# resp.input_config.data_source.s3_data_source.manifest_s3_uri #=> String
# resp.input_config.data_source.sns_data_source.sns_topic_arn #=> String
# resp.input_config.data_attributes.content_classifiers #=> Array
# resp.input_config.data_attributes.content_classifiers[0] #=> String, one of "FreeOfPersonallyIdentifiableInformation", "FreeOfAdultContent"
# resp.output_config.s3_output_path #=> String
# resp.output_config.kms_key_id #=> String
# resp.output_config.sns_topic_arn #=> String
# resp.role_arn #=> String
# resp.label_category_config_s3_uri #=> String
# resp.stopping_conditions.max_human_labeled_object_count #=> Integer
# resp.stopping_conditions.max_percentage_of_input_dataset_labeled #=> Integer
# resp.labeling_job_algorithms_config.labeling_job_algorithm_specification_arn #=> String
# resp.labeling_job_algorithms_config.initial_active_learning_model_arn #=> String
# resp.labeling_job_algorithms_config.labeling_job_resource_config.volume_kms_key_id #=> String
# resp.labeling_job_algorithms_config.labeling_job_resource_config.vpc_config.security_group_ids #=> Array
# resp.labeling_job_algorithms_config.labeling_job_resource_config.vpc_config.security_group_ids[0] #=> String
# resp.labeling_job_algorithms_config.labeling_job_resource_config.vpc_config.subnets #=> Array
# resp.labeling_job_algorithms_config.labeling_job_resource_config.vpc_config.subnets[0] #=> String
# resp.human_task_config.workteam_arn #=> String
# resp.human_task_config.ui_config.ui_template_s3_uri #=> String
# resp.human_task_config.ui_config.human_task_ui_arn #=> String
# resp.human_task_config.pre_human_task_lambda_arn #=> String
# resp.human_task_config.task_keywords #=> Array
# resp.human_task_config.task_keywords[0] #=> String
# resp.human_task_config.task_title #=> String
# resp.human_task_config.task_description #=> String
# resp.human_task_config.number_of_human_workers_per_data_object #=> Integer
# resp.human_task_config.task_time_limit_in_seconds #=> Integer
# resp.human_task_config.task_availability_lifetime_in_seconds #=> Integer
# resp.human_task_config.max_concurrent_task_count #=> Integer
# resp.human_task_config.annotation_consolidation_config.annotation_consolidation_lambda_arn #=> String
# resp.human_task_config.public_workforce_task_price.amount_in_usd.dollars #=> Integer
# resp.human_task_config.public_workforce_task_price.amount_in_usd.cents #=> Integer
# resp.human_task_config.public_workforce_task_price.amount_in_usd.tenth_fractions_of_a_cent #=> Integer
# resp.tags #=> Array
# resp.tags[0].key #=> String
# resp.tags[0].value #=> String
# resp.labeling_job_output.output_dataset_s3_uri #=> String
# resp.labeling_job_output.final_active_learning_model_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeLabelingJob AWS API Documentation
#
# @overload describe_labeling_job(params = {})
# @param [Hash] params ({})
def describe_labeling_job(params = {}, options = {})
req = build_request(:describe_labeling_job, params)
req.send_request(options)
end
# Provides a list of properties for the requested lineage group. For
# more information, see [ Cross-Account Lineage Tracking ][1] in the
# *Amazon SageMaker Developer Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/xaccount-lineage-tracking.html
#
# @option params [required, String] :lineage_group_name
# The name of the lineage group.
#
# @return [Types::DescribeLineageGroupResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeLineageGroupResponse#lineage_group_name #lineage_group_name} => String
# * {Types::DescribeLineageGroupResponse#lineage_group_arn #lineage_group_arn} => String
# * {Types::DescribeLineageGroupResponse#display_name #display_name} => String
# * {Types::DescribeLineageGroupResponse#description #description} => String
# * {Types::DescribeLineageGroupResponse#creation_time #creation_time} => Time
# * {Types::DescribeLineageGroupResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeLineageGroupResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeLineageGroupResponse#last_modified_by #last_modified_by} => Types::UserContext
#
# @example Request syntax with placeholder values
#
# resp = client.describe_lineage_group({
# lineage_group_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.lineage_group_name #=> String
# resp.lineage_group_arn #=> String
# resp.display_name #=> String
# resp.description #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeLineageGroup AWS API Documentation
#
# @overload describe_lineage_group(params = {})
# @param [Hash] params ({})
def describe_lineage_group(params = {}, options = {})
req = build_request(:describe_lineage_group, params)
req.send_request(options)
end
# Describes a model that you created using the `CreateModel` API.
#
# @option params [required, String] :model_name
# The name of the model.
#
# @return [Types::DescribeModelOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelOutput#model_name #model_name} => String
# * {Types::DescribeModelOutput#primary_container #primary_container} => Types::ContainerDefinition
# * {Types::DescribeModelOutput#containers #containers} => Array<Types::ContainerDefinition>
# * {Types::DescribeModelOutput#inference_execution_config #inference_execution_config} => Types::InferenceExecutionConfig
# * {Types::DescribeModelOutput#execution_role_arn #execution_role_arn} => String
# * {Types::DescribeModelOutput#vpc_config #vpc_config} => Types::VpcConfig
# * {Types::DescribeModelOutput#creation_time #creation_time} => Time
# * {Types::DescribeModelOutput#model_arn #model_arn} => String
# * {Types::DescribeModelOutput#enable_network_isolation #enable_network_isolation} => Boolean
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model({
# model_name: "ModelName", # required
# })
#
# @example Response structure
#
# resp.model_name #=> String
# resp.primary_container.container_hostname #=> String
# resp.primary_container.image #=> String
# resp.primary_container.image_config.repository_access_mode #=> String, one of "Platform", "Vpc"
# resp.primary_container.image_config.repository_auth_config.repository_credentials_provider_arn #=> String
# resp.primary_container.mode #=> String, one of "SingleModel", "MultiModel"
# resp.primary_container.model_data_url #=> String
# resp.primary_container.environment #=> Hash
# resp.primary_container.environment["EnvironmentKey"] #=> String
# resp.primary_container.model_package_name #=> String
# resp.primary_container.inference_specification_name #=> String
# resp.primary_container.multi_model_config.model_cache_setting #=> String, one of "Enabled", "Disabled"
# resp.containers #=> Array
# resp.containers[0].container_hostname #=> String
# resp.containers[0].image #=> String
# resp.containers[0].image_config.repository_access_mode #=> String, one of "Platform", "Vpc"
# resp.containers[0].image_config.repository_auth_config.repository_credentials_provider_arn #=> String
# resp.containers[0].mode #=> String, one of "SingleModel", "MultiModel"
# resp.containers[0].model_data_url #=> String
# resp.containers[0].environment #=> Hash
# resp.containers[0].environment["EnvironmentKey"] #=> String
# resp.containers[0].model_package_name #=> String
# resp.containers[0].inference_specification_name #=> String
# resp.containers[0].multi_model_config.model_cache_setting #=> String, one of "Enabled", "Disabled"
# resp.inference_execution_config.mode #=> String, one of "Serial", "Direct"
# resp.execution_role_arn #=> String
# resp.vpc_config.security_group_ids #=> Array
# resp.vpc_config.security_group_ids[0] #=> String
# resp.vpc_config.subnets #=> Array
# resp.vpc_config.subnets[0] #=> String
# resp.creation_time #=> Time
# resp.model_arn #=> String
# resp.enable_network_isolation #=> Boolean
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModel AWS API Documentation
#
# @overload describe_model(params = {})
# @param [Hash] params ({})
def describe_model(params = {}, options = {})
req = build_request(:describe_model, params)
req.send_request(options)
end
# Returns a description of a model bias job definition.
#
# @option params [required, String] :job_definition_name
# The name of the model bias job definition. The name must be unique
# within an Amazon Web Services Region in the Amazon Web Services
# account.
#
# @return [Types::DescribeModelBiasJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelBiasJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
# * {Types::DescribeModelBiasJobDefinitionResponse#job_definition_name #job_definition_name} => String
# * {Types::DescribeModelBiasJobDefinitionResponse#creation_time #creation_time} => Time
# * {Types::DescribeModelBiasJobDefinitionResponse#model_bias_baseline_config #model_bias_baseline_config} => Types::ModelBiasBaselineConfig
# * {Types::DescribeModelBiasJobDefinitionResponse#model_bias_app_specification #model_bias_app_specification} => Types::ModelBiasAppSpecification
# * {Types::DescribeModelBiasJobDefinitionResponse#model_bias_job_input #model_bias_job_input} => Types::ModelBiasJobInput
# * {Types::DescribeModelBiasJobDefinitionResponse#model_bias_job_output_config #model_bias_job_output_config} => Types::MonitoringOutputConfig
# * {Types::DescribeModelBiasJobDefinitionResponse#job_resources #job_resources} => Types::MonitoringResources
# * {Types::DescribeModelBiasJobDefinitionResponse#network_config #network_config} => Types::MonitoringNetworkConfig
# * {Types::DescribeModelBiasJobDefinitionResponse#role_arn #role_arn} => String
# * {Types::DescribeModelBiasJobDefinitionResponse#stopping_condition #stopping_condition} => Types::MonitoringStoppingCondition
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_bias_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
# resp.job_definition_name #=> String
# resp.creation_time #=> Time
# resp.model_bias_baseline_config.baselining_job_name #=> String
# resp.model_bias_baseline_config.constraints_resource.s3_uri #=> String
# resp.model_bias_app_specification.image_uri #=> String
# resp.model_bias_app_specification.config_uri #=> String
# resp.model_bias_app_specification.environment #=> Hash
# resp.model_bias_app_specification.environment["ProcessingEnvironmentKey"] #=> String
# resp.model_bias_job_input.endpoint_input.endpoint_name #=> String
# resp.model_bias_job_input.endpoint_input.local_path #=> String
# resp.model_bias_job_input.endpoint_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.model_bias_job_input.endpoint_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.model_bias_job_input.endpoint_input.features_attribute #=> String
# resp.model_bias_job_input.endpoint_input.inference_attribute #=> String
# resp.model_bias_job_input.endpoint_input.probability_attribute #=> String
# resp.model_bias_job_input.endpoint_input.probability_threshold_attribute #=> Float
# resp.model_bias_job_input.endpoint_input.start_time_offset #=> String
# resp.model_bias_job_input.endpoint_input.end_time_offset #=> String
# resp.model_bias_job_input.batch_transform_input.data_captured_destination_s3_uri #=> String
# resp.model_bias_job_input.batch_transform_input.dataset_format.csv.header #=> Boolean
# resp.model_bias_job_input.batch_transform_input.dataset_format.json.line #=> Boolean
# resp.model_bias_job_input.batch_transform_input.local_path #=> String
# resp.model_bias_job_input.batch_transform_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.model_bias_job_input.batch_transform_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.model_bias_job_input.batch_transform_input.features_attribute #=> String
# resp.model_bias_job_input.batch_transform_input.inference_attribute #=> String
# resp.model_bias_job_input.batch_transform_input.probability_attribute #=> String
# resp.model_bias_job_input.batch_transform_input.probability_threshold_attribute #=> Float
# resp.model_bias_job_input.batch_transform_input.start_time_offset #=> String
# resp.model_bias_job_input.batch_transform_input.end_time_offset #=> String
# resp.model_bias_job_input.ground_truth_s3_input.s3_uri #=> String
# resp.model_bias_job_output_config.monitoring_outputs #=> Array
# resp.model_bias_job_output_config.monitoring_outputs[0].s3_output.s3_uri #=> String
# resp.model_bias_job_output_config.monitoring_outputs[0].s3_output.local_path #=> String
# resp.model_bias_job_output_config.monitoring_outputs[0].s3_output.s3_upload_mode #=> String, one of "Continuous", "EndOfJob"
# resp.model_bias_job_output_config.kms_key_id #=> String
# resp.job_resources.cluster_config.instance_count #=> Integer
# resp.job_resources.cluster_config.instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.job_resources.cluster_config.volume_size_in_gb #=> Integer
# resp.job_resources.cluster_config.volume_kms_key_id #=> String
# resp.network_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.network_config.enable_network_isolation #=> Boolean
# resp.network_config.vpc_config.security_group_ids #=> Array
# resp.network_config.vpc_config.security_group_ids[0] #=> String
# resp.network_config.vpc_config.subnets #=> Array
# resp.network_config.vpc_config.subnets[0] #=> String
# resp.role_arn #=> String
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelBiasJobDefinition AWS API Documentation
#
# @overload describe_model_bias_job_definition(params = {})
# @param [Hash] params ({})
def describe_model_bias_job_definition(params = {}, options = {})
req = build_request(:describe_model_bias_job_definition, params)
req.send_request(options)
end
# Describes the content, creation time, and security configuration of an
# Amazon SageMaker Model Card.
#
# @option params [required, String] :model_card_name
# The name of the model card to describe.
#
# @option params [Integer] :model_card_version
# The version of the model card to describe. If a version is not
# provided, then the latest version of the model card is described.
#
# @return [Types::DescribeModelCardResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelCardResponse#model_card_arn #model_card_arn} => String
# * {Types::DescribeModelCardResponse#model_card_name #model_card_name} => String
# * {Types::DescribeModelCardResponse#model_card_version #model_card_version} => Integer
# * {Types::DescribeModelCardResponse#content #content} => String
# * {Types::DescribeModelCardResponse#model_card_status #model_card_status} => String
# * {Types::DescribeModelCardResponse#security_config #security_config} => Types::ModelCardSecurityConfig
# * {Types::DescribeModelCardResponse#creation_time #creation_time} => Time
# * {Types::DescribeModelCardResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeModelCardResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeModelCardResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeModelCardResponse#model_card_processing_status #model_card_processing_status} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_card({
# model_card_name: "EntityName", # required
# model_card_version: 1,
# })
#
# @example Response structure
#
# resp.model_card_arn #=> String
# resp.model_card_name #=> String
# resp.model_card_version #=> Integer
# resp.content #=> String
# resp.model_card_status #=> String, one of "Draft", "PendingReview", "Approved", "Archived"
# resp.security_config.kms_key_id #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.model_card_processing_status #=> String, one of "DeleteInProgress", "DeletePending", "ContentDeleted", "ExportJobsDeleted", "DeleteCompleted", "DeleteFailed"
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelCard AWS API Documentation
#
# @overload describe_model_card(params = {})
# @param [Hash] params ({})
def describe_model_card(params = {}, options = {})
req = build_request(:describe_model_card, params)
req.send_request(options)
end
# Describes an Amazon SageMaker Model Card export job.
#
# @option params [required, String] :model_card_export_job_arn
# The Amazon Resource Name (ARN) of the model card export job to
# describe.
#
# @return [Types::DescribeModelCardExportJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelCardExportJobResponse#model_card_export_job_name #model_card_export_job_name} => String
# * {Types::DescribeModelCardExportJobResponse#model_card_export_job_arn #model_card_export_job_arn} => String
# * {Types::DescribeModelCardExportJobResponse#status #status} => String
# * {Types::DescribeModelCardExportJobResponse#model_card_name #model_card_name} => String
# * {Types::DescribeModelCardExportJobResponse#model_card_version #model_card_version} => Integer
# * {Types::DescribeModelCardExportJobResponse#output_config #output_config} => Types::ModelCardExportOutputConfig
# * {Types::DescribeModelCardExportJobResponse#created_at #created_at} => Time
# * {Types::DescribeModelCardExportJobResponse#last_modified_at #last_modified_at} => Time
# * {Types::DescribeModelCardExportJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeModelCardExportJobResponse#export_artifacts #export_artifacts} => Types::ModelCardExportArtifacts
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_card_export_job({
# model_card_export_job_arn: "ModelCardExportJobArn", # required
# })
#
# @example Response structure
#
# resp.model_card_export_job_name #=> String
# resp.model_card_export_job_arn #=> String
# resp.status #=> String, one of "InProgress", "Completed", "Failed"
# resp.model_card_name #=> String
# resp.model_card_version #=> Integer
# resp.output_config.s3_output_path #=> String
# resp.created_at #=> Time
# resp.last_modified_at #=> Time
# resp.failure_reason #=> String
# resp.export_artifacts.s3_export_artifacts #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelCardExportJob AWS API Documentation
#
# @overload describe_model_card_export_job(params = {})
# @param [Hash] params ({})
def describe_model_card_export_job(params = {}, options = {})
req = build_request(:describe_model_card_export_job, params)
req.send_request(options)
end
# Returns a description of a model explainability job definition.
#
# @option params [required, String] :job_definition_name
# The name of the model explainability job definition. The name must be
# unique within an Amazon Web Services Region in the Amazon Web Services
# account.
#
# @return [Types::DescribeModelExplainabilityJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#job_definition_name #job_definition_name} => String
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#creation_time #creation_time} => Time
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#model_explainability_baseline_config #model_explainability_baseline_config} => Types::ModelExplainabilityBaselineConfig
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#model_explainability_app_specification #model_explainability_app_specification} => Types::ModelExplainabilityAppSpecification
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#model_explainability_job_input #model_explainability_job_input} => Types::ModelExplainabilityJobInput
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#model_explainability_job_output_config #model_explainability_job_output_config} => Types::MonitoringOutputConfig
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#job_resources #job_resources} => Types::MonitoringResources
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#network_config #network_config} => Types::MonitoringNetworkConfig
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#role_arn #role_arn} => String
# * {Types::DescribeModelExplainabilityJobDefinitionResponse#stopping_condition #stopping_condition} => Types::MonitoringStoppingCondition
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_explainability_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
# resp.job_definition_name #=> String
# resp.creation_time #=> Time
# resp.model_explainability_baseline_config.baselining_job_name #=> String
# resp.model_explainability_baseline_config.constraints_resource.s3_uri #=> String
# resp.model_explainability_app_specification.image_uri #=> String
# resp.model_explainability_app_specification.config_uri #=> String
# resp.model_explainability_app_specification.environment #=> Hash
# resp.model_explainability_app_specification.environment["ProcessingEnvironmentKey"] #=> String
# resp.model_explainability_job_input.endpoint_input.endpoint_name #=> String
# resp.model_explainability_job_input.endpoint_input.local_path #=> String
# resp.model_explainability_job_input.endpoint_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.model_explainability_job_input.endpoint_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.model_explainability_job_input.endpoint_input.features_attribute #=> String
# resp.model_explainability_job_input.endpoint_input.inference_attribute #=> String
# resp.model_explainability_job_input.endpoint_input.probability_attribute #=> String
# resp.model_explainability_job_input.endpoint_input.probability_threshold_attribute #=> Float
# resp.model_explainability_job_input.endpoint_input.start_time_offset #=> String
# resp.model_explainability_job_input.endpoint_input.end_time_offset #=> String
# resp.model_explainability_job_input.batch_transform_input.data_captured_destination_s3_uri #=> String
# resp.model_explainability_job_input.batch_transform_input.dataset_format.csv.header #=> Boolean
# resp.model_explainability_job_input.batch_transform_input.dataset_format.json.line #=> Boolean
# resp.model_explainability_job_input.batch_transform_input.local_path #=> String
# resp.model_explainability_job_input.batch_transform_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.model_explainability_job_input.batch_transform_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.model_explainability_job_input.batch_transform_input.features_attribute #=> String
# resp.model_explainability_job_input.batch_transform_input.inference_attribute #=> String
# resp.model_explainability_job_input.batch_transform_input.probability_attribute #=> String
# resp.model_explainability_job_input.batch_transform_input.probability_threshold_attribute #=> Float
# resp.model_explainability_job_input.batch_transform_input.start_time_offset #=> String
# resp.model_explainability_job_input.batch_transform_input.end_time_offset #=> String
# resp.model_explainability_job_output_config.monitoring_outputs #=> Array
# resp.model_explainability_job_output_config.monitoring_outputs[0].s3_output.s3_uri #=> String
# resp.model_explainability_job_output_config.monitoring_outputs[0].s3_output.local_path #=> String
# resp.model_explainability_job_output_config.monitoring_outputs[0].s3_output.s3_upload_mode #=> String, one of "Continuous", "EndOfJob"
# resp.model_explainability_job_output_config.kms_key_id #=> String
# resp.job_resources.cluster_config.instance_count #=> Integer
# resp.job_resources.cluster_config.instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.job_resources.cluster_config.volume_size_in_gb #=> Integer
# resp.job_resources.cluster_config.volume_kms_key_id #=> String
# resp.network_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.network_config.enable_network_isolation #=> Boolean
# resp.network_config.vpc_config.security_group_ids #=> Array
# resp.network_config.vpc_config.security_group_ids[0] #=> String
# resp.network_config.vpc_config.subnets #=> Array
# resp.network_config.vpc_config.subnets[0] #=> String
# resp.role_arn #=> String
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelExplainabilityJobDefinition AWS API Documentation
#
# @overload describe_model_explainability_job_definition(params = {})
# @param [Hash] params ({})
def describe_model_explainability_job_definition(params = {}, options = {})
req = build_request(:describe_model_explainability_job_definition, params)
req.send_request(options)
end
# Returns a description of the specified model package, which is used to
# create SageMaker models or list them on Amazon Web Services
# Marketplace.
#
# To create models in SageMaker, buyers can subscribe to model packages
# listed on Amazon Web Services Marketplace.
#
# @option params [required, String] :model_package_name
# The name or Amazon Resource Name (ARN) of the model package to
# describe.
#
# When you specify a name, the name must have 1 to 63 characters. Valid
# characters are a-z, A-Z, 0-9, and - (hyphen).
#
# @return [Types::DescribeModelPackageOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelPackageOutput#model_package_name #model_package_name} => String
# * {Types::DescribeModelPackageOutput#model_package_group_name #model_package_group_name} => String
# * {Types::DescribeModelPackageOutput#model_package_version #model_package_version} => Integer
# * {Types::DescribeModelPackageOutput#model_package_arn #model_package_arn} => String
# * {Types::DescribeModelPackageOutput#model_package_description #model_package_description} => String
# * {Types::DescribeModelPackageOutput#creation_time #creation_time} => Time
# * {Types::DescribeModelPackageOutput#inference_specification #inference_specification} => Types::InferenceSpecification
# * {Types::DescribeModelPackageOutput#source_algorithm_specification #source_algorithm_specification} => Types::SourceAlgorithmSpecification
# * {Types::DescribeModelPackageOutput#validation_specification #validation_specification} => Types::ModelPackageValidationSpecification
# * {Types::DescribeModelPackageOutput#model_package_status #model_package_status} => String
# * {Types::DescribeModelPackageOutput#model_package_status_details #model_package_status_details} => Types::ModelPackageStatusDetails
# * {Types::DescribeModelPackageOutput#certify_for_marketplace #certify_for_marketplace} => Boolean
# * {Types::DescribeModelPackageOutput#model_approval_status #model_approval_status} => String
# * {Types::DescribeModelPackageOutput#created_by #created_by} => Types::UserContext
# * {Types::DescribeModelPackageOutput#metadata_properties #metadata_properties} => Types::MetadataProperties
# * {Types::DescribeModelPackageOutput#model_metrics #model_metrics} => Types::ModelMetrics
# * {Types::DescribeModelPackageOutput#last_modified_time #last_modified_time} => Time
# * {Types::DescribeModelPackageOutput#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeModelPackageOutput#approval_description #approval_description} => String
# * {Types::DescribeModelPackageOutput#customer_metadata_properties #customer_metadata_properties} => Hash<String,String>
# * {Types::DescribeModelPackageOutput#drift_check_baselines #drift_check_baselines} => Types::DriftCheckBaselines
# * {Types::DescribeModelPackageOutput#domain #domain} => String
# * {Types::DescribeModelPackageOutput#task #task} => String
# * {Types::DescribeModelPackageOutput#sample_payload_url #sample_payload_url} => String
# * {Types::DescribeModelPackageOutput#additional_inference_specifications #additional_inference_specifications} => Array<Types::AdditionalInferenceSpecificationDefinition>
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_package({
# model_package_name: "VersionedArnOrName", # required
# })
#
# @example Response structure
#
# resp.model_package_name #=> String
# resp.model_package_group_name #=> String
# resp.model_package_version #=> Integer
# resp.model_package_arn #=> String
# resp.model_package_description #=> String
# resp.creation_time #=> Time
# resp.inference_specification.containers #=> Array
# resp.inference_specification.containers[0].container_hostname #=> String
# resp.inference_specification.containers[0].image #=> String
# resp.inference_specification.containers[0].image_digest #=> String
# resp.inference_specification.containers[0].model_data_url #=> String
# resp.inference_specification.containers[0].product_id #=> String
# resp.inference_specification.containers[0].environment #=> Hash
# resp.inference_specification.containers[0].environment["EnvironmentKey"] #=> String
# resp.inference_specification.containers[0].model_input.data_input_config #=> String
# resp.inference_specification.containers[0].framework #=> String
# resp.inference_specification.containers[0].framework_version #=> String
# resp.inference_specification.containers[0].nearest_model_name #=> String
# resp.inference_specification.supported_transform_instance_types #=> Array
# resp.inference_specification.supported_transform_instance_types[0] #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.inference_specification.supported_realtime_inference_instance_types #=> Array
# resp.inference_specification.supported_realtime_inference_instance_types[0] #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.inference_specification.supported_content_types #=> Array
# resp.inference_specification.supported_content_types[0] #=> String
# resp.inference_specification.supported_response_mime_types #=> Array
# resp.inference_specification.supported_response_mime_types[0] #=> String
# resp.source_algorithm_specification.source_algorithms #=> Array
# resp.source_algorithm_specification.source_algorithms[0].model_data_url #=> String
# resp.source_algorithm_specification.source_algorithms[0].algorithm_name #=> String
# resp.validation_specification.validation_role #=> String
# resp.validation_specification.validation_profiles #=> Array
# resp.validation_specification.validation_profiles[0].profile_name #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.max_concurrent_transforms #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.max_payload_in_mb #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.batch_strategy #=> String, one of "MultiRecord", "SingleRecord"
# resp.validation_specification.validation_profiles[0].transform_job_definition.environment #=> Hash
# resp.validation_specification.validation_profiles[0].transform_job_definition.environment["TransformEnvironmentKey"] #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.data_source.s3_data_source.s3_uri #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.content_type #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.compression_type #=> String, one of "None", "Gzip"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_input.split_type #=> String, one of "None", "Line", "RecordIO", "TFRecord"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.s3_output_path #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.accept #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.assemble_with #=> String, one of "None", "Line"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_output.kms_key_id #=> String
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_resources.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_resources.instance_count #=> Integer
# resp.validation_specification.validation_profiles[0].transform_job_definition.transform_resources.volume_kms_key_id #=> String
# resp.model_package_status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Deleting"
# resp.model_package_status_details.validation_statuses #=> Array
# resp.model_package_status_details.validation_statuses[0].name #=> String
# resp.model_package_status_details.validation_statuses[0].status #=> String, one of "NotStarted", "InProgress", "Completed", "Failed"
# resp.model_package_status_details.validation_statuses[0].failure_reason #=> String
# resp.model_package_status_details.image_scan_statuses #=> Array
# resp.model_package_status_details.image_scan_statuses[0].name #=> String
# resp.model_package_status_details.image_scan_statuses[0].status #=> String, one of "NotStarted", "InProgress", "Completed", "Failed"
# resp.model_package_status_details.image_scan_statuses[0].failure_reason #=> String
# resp.certify_for_marketplace #=> Boolean
# resp.model_approval_status #=> String, one of "Approved", "Rejected", "PendingManualApproval"
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.metadata_properties.commit_id #=> String
# resp.metadata_properties.repository #=> String
# resp.metadata_properties.generated_by #=> String
# resp.metadata_properties.project_id #=> String
# resp.model_metrics.model_quality.statistics.content_type #=> String
# resp.model_metrics.model_quality.statistics.content_digest #=> String
# resp.model_metrics.model_quality.statistics.s3_uri #=> String
# resp.model_metrics.model_quality.constraints.content_type #=> String
# resp.model_metrics.model_quality.constraints.content_digest #=> String
# resp.model_metrics.model_quality.constraints.s3_uri #=> String
# resp.model_metrics.model_data_quality.statistics.content_type #=> String
# resp.model_metrics.model_data_quality.statistics.content_digest #=> String
# resp.model_metrics.model_data_quality.statistics.s3_uri #=> String
# resp.model_metrics.model_data_quality.constraints.content_type #=> String
# resp.model_metrics.model_data_quality.constraints.content_digest #=> String
# resp.model_metrics.model_data_quality.constraints.s3_uri #=> String
# resp.model_metrics.bias.report.content_type #=> String
# resp.model_metrics.bias.report.content_digest #=> String
# resp.model_metrics.bias.report.s3_uri #=> String
# resp.model_metrics.bias.pre_training_report.content_type #=> String
# resp.model_metrics.bias.pre_training_report.content_digest #=> String
# resp.model_metrics.bias.pre_training_report.s3_uri #=> String
# resp.model_metrics.bias.post_training_report.content_type #=> String
# resp.model_metrics.bias.post_training_report.content_digest #=> String
# resp.model_metrics.bias.post_training_report.s3_uri #=> String
# resp.model_metrics.explainability.report.content_type #=> String
# resp.model_metrics.explainability.report.content_digest #=> String
# resp.model_metrics.explainability.report.s3_uri #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.approval_description #=> String
# resp.customer_metadata_properties #=> Hash
# resp.customer_metadata_properties["CustomerMetadataKey"] #=> String
# resp.drift_check_baselines.bias.config_file.content_type #=> String
# resp.drift_check_baselines.bias.config_file.content_digest #=> String
# resp.drift_check_baselines.bias.config_file.s3_uri #=> String
# resp.drift_check_baselines.bias.pre_training_constraints.content_type #=> String
# resp.drift_check_baselines.bias.pre_training_constraints.content_digest #=> String
# resp.drift_check_baselines.bias.pre_training_constraints.s3_uri #=> String
# resp.drift_check_baselines.bias.post_training_constraints.content_type #=> String
# resp.drift_check_baselines.bias.post_training_constraints.content_digest #=> String
# resp.drift_check_baselines.bias.post_training_constraints.s3_uri #=> String
# resp.drift_check_baselines.explainability.constraints.content_type #=> String
# resp.drift_check_baselines.explainability.constraints.content_digest #=> String
# resp.drift_check_baselines.explainability.constraints.s3_uri #=> String
# resp.drift_check_baselines.explainability.config_file.content_type #=> String
# resp.drift_check_baselines.explainability.config_file.content_digest #=> String
# resp.drift_check_baselines.explainability.config_file.s3_uri #=> String
# resp.drift_check_baselines.model_quality.statistics.content_type #=> String
# resp.drift_check_baselines.model_quality.statistics.content_digest #=> String
# resp.drift_check_baselines.model_quality.statistics.s3_uri #=> String
# resp.drift_check_baselines.model_quality.constraints.content_type #=> String
# resp.drift_check_baselines.model_quality.constraints.content_digest #=> String
# resp.drift_check_baselines.model_quality.constraints.s3_uri #=> String
# resp.drift_check_baselines.model_data_quality.statistics.content_type #=> String
# resp.drift_check_baselines.model_data_quality.statistics.content_digest #=> String
# resp.drift_check_baselines.model_data_quality.statistics.s3_uri #=> String
# resp.drift_check_baselines.model_data_quality.constraints.content_type #=> String
# resp.drift_check_baselines.model_data_quality.constraints.content_digest #=> String
# resp.drift_check_baselines.model_data_quality.constraints.s3_uri #=> String
# resp.domain #=> String
# resp.task #=> String
# resp.sample_payload_url #=> String
# resp.additional_inference_specifications #=> Array
# resp.additional_inference_specifications[0].name #=> String
# resp.additional_inference_specifications[0].description #=> String
# resp.additional_inference_specifications[0].containers #=> Array
# resp.additional_inference_specifications[0].containers[0].container_hostname #=> String
# resp.additional_inference_specifications[0].containers[0].image #=> String
# resp.additional_inference_specifications[0].containers[0].image_digest #=> String
# resp.additional_inference_specifications[0].containers[0].model_data_url #=> String
# resp.additional_inference_specifications[0].containers[0].product_id #=> String
# resp.additional_inference_specifications[0].containers[0].environment #=> Hash
# resp.additional_inference_specifications[0].containers[0].environment["EnvironmentKey"] #=> String
# resp.additional_inference_specifications[0].containers[0].model_input.data_input_config #=> String
# resp.additional_inference_specifications[0].containers[0].framework #=> String
# resp.additional_inference_specifications[0].containers[0].framework_version #=> String
# resp.additional_inference_specifications[0].containers[0].nearest_model_name #=> String
# resp.additional_inference_specifications[0].supported_transform_instance_types #=> Array
# resp.additional_inference_specifications[0].supported_transform_instance_types[0] #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.additional_inference_specifications[0].supported_realtime_inference_instance_types #=> Array
# resp.additional_inference_specifications[0].supported_realtime_inference_instance_types[0] #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.additional_inference_specifications[0].supported_content_types #=> Array
# resp.additional_inference_specifications[0].supported_content_types[0] #=> String
# resp.additional_inference_specifications[0].supported_response_mime_types #=> Array
# resp.additional_inference_specifications[0].supported_response_mime_types[0] #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelPackage AWS API Documentation
#
# @overload describe_model_package(params = {})
# @param [Hash] params ({})
def describe_model_package(params = {}, options = {})
req = build_request(:describe_model_package, params)
req.send_request(options)
end
# Gets a description for the specified model group.
#
# @option params [required, String] :model_package_group_name
# The name of gthe model group to describe.
#
# @return [Types::DescribeModelPackageGroupOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelPackageGroupOutput#model_package_group_name #model_package_group_name} => String
# * {Types::DescribeModelPackageGroupOutput#model_package_group_arn #model_package_group_arn} => String
# * {Types::DescribeModelPackageGroupOutput#model_package_group_description #model_package_group_description} => String
# * {Types::DescribeModelPackageGroupOutput#creation_time #creation_time} => Time
# * {Types::DescribeModelPackageGroupOutput#created_by #created_by} => Types::UserContext
# * {Types::DescribeModelPackageGroupOutput#model_package_group_status #model_package_group_status} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_package_group({
# model_package_group_name: "ArnOrName", # required
# })
#
# @example Response structure
#
# resp.model_package_group_name #=> String
# resp.model_package_group_arn #=> String
# resp.model_package_group_description #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.model_package_group_status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Deleting", "DeleteFailed"
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelPackageGroup AWS API Documentation
#
# @overload describe_model_package_group(params = {})
# @param [Hash] params ({})
def describe_model_package_group(params = {}, options = {})
req = build_request(:describe_model_package_group, params)
req.send_request(options)
end
# Returns a description of a model quality job definition.
#
# @option params [required, String] :job_definition_name
# The name of the model quality job. The name must be unique within an
# Amazon Web Services Region in the Amazon Web Services account.
#
# @return [Types::DescribeModelQualityJobDefinitionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeModelQualityJobDefinitionResponse#job_definition_arn #job_definition_arn} => String
# * {Types::DescribeModelQualityJobDefinitionResponse#job_definition_name #job_definition_name} => String
# * {Types::DescribeModelQualityJobDefinitionResponse#creation_time #creation_time} => Time
# * {Types::DescribeModelQualityJobDefinitionResponse#model_quality_baseline_config #model_quality_baseline_config} => Types::ModelQualityBaselineConfig
# * {Types::DescribeModelQualityJobDefinitionResponse#model_quality_app_specification #model_quality_app_specification} => Types::ModelQualityAppSpecification
# * {Types::DescribeModelQualityJobDefinitionResponse#model_quality_job_input #model_quality_job_input} => Types::ModelQualityJobInput
# * {Types::DescribeModelQualityJobDefinitionResponse#model_quality_job_output_config #model_quality_job_output_config} => Types::MonitoringOutputConfig
# * {Types::DescribeModelQualityJobDefinitionResponse#job_resources #job_resources} => Types::MonitoringResources
# * {Types::DescribeModelQualityJobDefinitionResponse#network_config #network_config} => Types::MonitoringNetworkConfig
# * {Types::DescribeModelQualityJobDefinitionResponse#role_arn #role_arn} => String
# * {Types::DescribeModelQualityJobDefinitionResponse#stopping_condition #stopping_condition} => Types::MonitoringStoppingCondition
#
# @example Request syntax with placeholder values
#
# resp = client.describe_model_quality_job_definition({
# job_definition_name: "MonitoringJobDefinitionName", # required
# })
#
# @example Response structure
#
# resp.job_definition_arn #=> String
# resp.job_definition_name #=> String
# resp.creation_time #=> Time
# resp.model_quality_baseline_config.baselining_job_name #=> String
# resp.model_quality_baseline_config.constraints_resource.s3_uri #=> String
# resp.model_quality_app_specification.image_uri #=> String
# resp.model_quality_app_specification.container_entrypoint #=> Array
# resp.model_quality_app_specification.container_entrypoint[0] #=> String
# resp.model_quality_app_specification.container_arguments #=> Array
# resp.model_quality_app_specification.container_arguments[0] #=> String
# resp.model_quality_app_specification.record_preprocessor_source_uri #=> String
# resp.model_quality_app_specification.post_analytics_processor_source_uri #=> String
# resp.model_quality_app_specification.problem_type #=> String, one of "BinaryClassification", "MulticlassClassification", "Regression"
# resp.model_quality_app_specification.environment #=> Hash
# resp.model_quality_app_specification.environment["ProcessingEnvironmentKey"] #=> String
# resp.model_quality_job_input.endpoint_input.endpoint_name #=> String
# resp.model_quality_job_input.endpoint_input.local_path #=> String
# resp.model_quality_job_input.endpoint_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.model_quality_job_input.endpoint_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.model_quality_job_input.endpoint_input.features_attribute #=> String
# resp.model_quality_job_input.endpoint_input.inference_attribute #=> String
# resp.model_quality_job_input.endpoint_input.probability_attribute #=> String
# resp.model_quality_job_input.endpoint_input.probability_threshold_attribute #=> Float
# resp.model_quality_job_input.endpoint_input.start_time_offset #=> String
# resp.model_quality_job_input.endpoint_input.end_time_offset #=> String
# resp.model_quality_job_input.batch_transform_input.data_captured_destination_s3_uri #=> String
# resp.model_quality_job_input.batch_transform_input.dataset_format.csv.header #=> Boolean
# resp.model_quality_job_input.batch_transform_input.dataset_format.json.line #=> Boolean
# resp.model_quality_job_input.batch_transform_input.local_path #=> String
# resp.model_quality_job_input.batch_transform_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.model_quality_job_input.batch_transform_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.model_quality_job_input.batch_transform_input.features_attribute #=> String
# resp.model_quality_job_input.batch_transform_input.inference_attribute #=> String
# resp.model_quality_job_input.batch_transform_input.probability_attribute #=> String
# resp.model_quality_job_input.batch_transform_input.probability_threshold_attribute #=> Float
# resp.model_quality_job_input.batch_transform_input.start_time_offset #=> String
# resp.model_quality_job_input.batch_transform_input.end_time_offset #=> String
# resp.model_quality_job_input.ground_truth_s3_input.s3_uri #=> String
# resp.model_quality_job_output_config.monitoring_outputs #=> Array
# resp.model_quality_job_output_config.monitoring_outputs[0].s3_output.s3_uri #=> String
# resp.model_quality_job_output_config.monitoring_outputs[0].s3_output.local_path #=> String
# resp.model_quality_job_output_config.monitoring_outputs[0].s3_output.s3_upload_mode #=> String, one of "Continuous", "EndOfJob"
# resp.model_quality_job_output_config.kms_key_id #=> String
# resp.job_resources.cluster_config.instance_count #=> Integer
# resp.job_resources.cluster_config.instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.job_resources.cluster_config.volume_size_in_gb #=> Integer
# resp.job_resources.cluster_config.volume_kms_key_id #=> String
# resp.network_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.network_config.enable_network_isolation #=> Boolean
# resp.network_config.vpc_config.security_group_ids #=> Array
# resp.network_config.vpc_config.security_group_ids[0] #=> String
# resp.network_config.vpc_config.subnets #=> Array
# resp.network_config.vpc_config.subnets[0] #=> String
# resp.role_arn #=> String
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeModelQualityJobDefinition AWS API Documentation
#
# @overload describe_model_quality_job_definition(params = {})
# @param [Hash] params ({})
def describe_model_quality_job_definition(params = {}, options = {})
req = build_request(:describe_model_quality_job_definition, params)
req.send_request(options)
end
# Describes the schedule for a monitoring job.
#
# @option params [required, String] :monitoring_schedule_name
# Name of a previously created monitoring schedule.
#
# @return [Types::DescribeMonitoringScheduleResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeMonitoringScheduleResponse#monitoring_schedule_arn #monitoring_schedule_arn} => String
# * {Types::DescribeMonitoringScheduleResponse#monitoring_schedule_name #monitoring_schedule_name} => String
# * {Types::DescribeMonitoringScheduleResponse#monitoring_schedule_status #monitoring_schedule_status} => String
# * {Types::DescribeMonitoringScheduleResponse#monitoring_type #monitoring_type} => String
# * {Types::DescribeMonitoringScheduleResponse#failure_reason #failure_reason} => String
# * {Types::DescribeMonitoringScheduleResponse#creation_time #creation_time} => Time
# * {Types::DescribeMonitoringScheduleResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeMonitoringScheduleResponse#monitoring_schedule_config #monitoring_schedule_config} => Types::MonitoringScheduleConfig
# * {Types::DescribeMonitoringScheduleResponse#endpoint_name #endpoint_name} => String
# * {Types::DescribeMonitoringScheduleResponse#last_monitoring_execution_summary #last_monitoring_execution_summary} => Types::MonitoringExecutionSummary
#
# @example Request syntax with placeholder values
#
# resp = client.describe_monitoring_schedule({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# })
#
# @example Response structure
#
# resp.monitoring_schedule_arn #=> String
# resp.monitoring_schedule_name #=> String
# resp.monitoring_schedule_status #=> String, one of "Pending", "Failed", "Scheduled", "Stopped"
# resp.monitoring_type #=> String, one of "DataQuality", "ModelQuality", "ModelBias", "ModelExplainability"
# resp.failure_reason #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.monitoring_schedule_config.schedule_config.schedule_expression #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.baseline_config.baselining_job_name #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.baseline_config.constraints_resource.s3_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.baseline_config.statistics_resource.s3_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs #=> Array
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.endpoint_name #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.local_path #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.features_attribute #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.inference_attribute #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.probability_attribute #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.probability_threshold_attribute #=> Float
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.start_time_offset #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].endpoint_input.end_time_offset #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.data_captured_destination_s3_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.dataset_format.csv.header #=> Boolean
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.dataset_format.json.line #=> Boolean
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.local_path #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.features_attribute #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.inference_attribute #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.probability_attribute #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.probability_threshold_attribute #=> Float
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.start_time_offset #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_inputs[0].batch_transform_input.end_time_offset #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_output_config.monitoring_outputs #=> Array
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_output_config.monitoring_outputs[0].s3_output.s3_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_output_config.monitoring_outputs[0].s3_output.local_path #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_output_config.monitoring_outputs[0].s3_output.s3_upload_mode #=> String, one of "Continuous", "EndOfJob"
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_output_config.kms_key_id #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_resources.cluster_config.instance_count #=> Integer
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_resources.cluster_config.instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_resources.cluster_config.volume_size_in_gb #=> Integer
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_resources.cluster_config.volume_kms_key_id #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.image_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.container_entrypoint #=> Array
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.container_entrypoint[0] #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.container_arguments #=> Array
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.container_arguments[0] #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.record_preprocessor_source_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.monitoring_app_specification.post_analytics_processor_source_uri #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.stopping_condition.max_runtime_in_seconds #=> Integer
# resp.monitoring_schedule_config.monitoring_job_definition.environment #=> Hash
# resp.monitoring_schedule_config.monitoring_job_definition.environment["ProcessingEnvironmentKey"] #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.network_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.monitoring_schedule_config.monitoring_job_definition.network_config.enable_network_isolation #=> Boolean
# resp.monitoring_schedule_config.monitoring_job_definition.network_config.vpc_config.security_group_ids #=> Array
# resp.monitoring_schedule_config.monitoring_job_definition.network_config.vpc_config.security_group_ids[0] #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.network_config.vpc_config.subnets #=> Array
# resp.monitoring_schedule_config.monitoring_job_definition.network_config.vpc_config.subnets[0] #=> String
# resp.monitoring_schedule_config.monitoring_job_definition.role_arn #=> String
# resp.monitoring_schedule_config.monitoring_job_definition_name #=> String
# resp.monitoring_schedule_config.monitoring_type #=> String, one of "DataQuality", "ModelQuality", "ModelBias", "ModelExplainability"
# resp.endpoint_name #=> String
# resp.last_monitoring_execution_summary.monitoring_schedule_name #=> String
# resp.last_monitoring_execution_summary.scheduled_time #=> Time
# resp.last_monitoring_execution_summary.creation_time #=> Time
# resp.last_monitoring_execution_summary.last_modified_time #=> Time
# resp.last_monitoring_execution_summary.monitoring_execution_status #=> String, one of "Pending", "Completed", "CompletedWithViolations", "InProgress", "Failed", "Stopping", "Stopped"
# resp.last_monitoring_execution_summary.processing_job_arn #=> String
# resp.last_monitoring_execution_summary.endpoint_name #=> String
# resp.last_monitoring_execution_summary.failure_reason #=> String
# resp.last_monitoring_execution_summary.monitoring_job_definition_name #=> String
# resp.last_monitoring_execution_summary.monitoring_type #=> String, one of "DataQuality", "ModelQuality", "ModelBias", "ModelExplainability"
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeMonitoringSchedule AWS API Documentation
#
# @overload describe_monitoring_schedule(params = {})
# @param [Hash] params ({})
def describe_monitoring_schedule(params = {}, options = {})
req = build_request(:describe_monitoring_schedule, params)
req.send_request(options)
end
# Returns information about a notebook instance.
#
# @option params [required, String] :notebook_instance_name
# The name of the notebook instance that you want information about.
#
# @return [Types::DescribeNotebookInstanceOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeNotebookInstanceOutput#notebook_instance_arn #notebook_instance_arn} => String
# * {Types::DescribeNotebookInstanceOutput#notebook_instance_name #notebook_instance_name} => String
# * {Types::DescribeNotebookInstanceOutput#notebook_instance_status #notebook_instance_status} => String
# * {Types::DescribeNotebookInstanceOutput#failure_reason #failure_reason} => String
# * {Types::DescribeNotebookInstanceOutput#url #url} => String
# * {Types::DescribeNotebookInstanceOutput#instance_type #instance_type} => String
# * {Types::DescribeNotebookInstanceOutput#subnet_id #subnet_id} => String
# * {Types::DescribeNotebookInstanceOutput#security_groups #security_groups} => Array<String>
# * {Types::DescribeNotebookInstanceOutput#role_arn #role_arn} => String
# * {Types::DescribeNotebookInstanceOutput#kms_key_id #kms_key_id} => String
# * {Types::DescribeNotebookInstanceOutput#network_interface_id #network_interface_id} => String
# * {Types::DescribeNotebookInstanceOutput#last_modified_time #last_modified_time} => Time
# * {Types::DescribeNotebookInstanceOutput#creation_time #creation_time} => Time
# * {Types::DescribeNotebookInstanceOutput#notebook_instance_lifecycle_config_name #notebook_instance_lifecycle_config_name} => String
# * {Types::DescribeNotebookInstanceOutput#direct_internet_access #direct_internet_access} => String
# * {Types::DescribeNotebookInstanceOutput#volume_size_in_gb #volume_size_in_gb} => Integer
# * {Types::DescribeNotebookInstanceOutput#accelerator_types #accelerator_types} => Array<String>
# * {Types::DescribeNotebookInstanceOutput#default_code_repository #default_code_repository} => String
# * {Types::DescribeNotebookInstanceOutput#additional_code_repositories #additional_code_repositories} => Array<String>
# * {Types::DescribeNotebookInstanceOutput#root_access #root_access} => String
# * {Types::DescribeNotebookInstanceOutput#platform_identifier #platform_identifier} => String
# * {Types::DescribeNotebookInstanceOutput#instance_metadata_service_configuration #instance_metadata_service_configuration} => Types::InstanceMetadataServiceConfiguration
#
# @example Request syntax with placeholder values
#
# resp = client.describe_notebook_instance({
# notebook_instance_name: "NotebookInstanceName", # required
# })
#
# @example Response structure
#
# resp.notebook_instance_arn #=> String
# resp.notebook_instance_name #=> String
# resp.notebook_instance_status #=> String, one of "Pending", "InService", "Stopping", "Stopped", "Failed", "Deleting", "Updating"
# resp.failure_reason #=> String
# resp.url #=> String
# resp.instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge"
# resp.subnet_id #=> String
# resp.security_groups #=> Array
# resp.security_groups[0] #=> String
# resp.role_arn #=> String
# resp.kms_key_id #=> String
# resp.network_interface_id #=> String
# resp.last_modified_time #=> Time
# resp.creation_time #=> Time
# resp.notebook_instance_lifecycle_config_name #=> String
# resp.direct_internet_access #=> String, one of "Enabled", "Disabled"
# resp.volume_size_in_gb #=> Integer
# resp.accelerator_types #=> Array
# resp.accelerator_types[0] #=> String, one of "ml.eia1.medium", "ml.eia1.large", "ml.eia1.xlarge", "ml.eia2.medium", "ml.eia2.large", "ml.eia2.xlarge"
# resp.default_code_repository #=> String
# resp.additional_code_repositories #=> Array
# resp.additional_code_repositories[0] #=> String
# resp.root_access #=> String, one of "Enabled", "Disabled"
# resp.platform_identifier #=> String
# resp.instance_metadata_service_configuration.minimum_instance_metadata_service_version #=> String
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * notebook_instance_deleted
# * notebook_instance_in_service
# * notebook_instance_stopped
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeNotebookInstance AWS API Documentation
#
# @overload describe_notebook_instance(params = {})
# @param [Hash] params ({})
def describe_notebook_instance(params = {}, options = {})
req = build_request(:describe_notebook_instance, params)
req.send_request(options)
end
# Returns a description of a notebook instance lifecycle configuration.
#
# For information about notebook instance lifestyle configurations, see
# [Step 2.1: (Optional) Customize a Notebook Instance][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/notebook-lifecycle-config.html
#
# @option params [required, String] :notebook_instance_lifecycle_config_name
# The name of the lifecycle configuration to describe.
#
# @return [Types::DescribeNotebookInstanceLifecycleConfigOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeNotebookInstanceLifecycleConfigOutput#notebook_instance_lifecycle_config_arn #notebook_instance_lifecycle_config_arn} => String
# * {Types::DescribeNotebookInstanceLifecycleConfigOutput#notebook_instance_lifecycle_config_name #notebook_instance_lifecycle_config_name} => String
# * {Types::DescribeNotebookInstanceLifecycleConfigOutput#on_create #on_create} => Array<Types::NotebookInstanceLifecycleHook>
# * {Types::DescribeNotebookInstanceLifecycleConfigOutput#on_start #on_start} => Array<Types::NotebookInstanceLifecycleHook>
# * {Types::DescribeNotebookInstanceLifecycleConfigOutput#last_modified_time #last_modified_time} => Time
# * {Types::DescribeNotebookInstanceLifecycleConfigOutput#creation_time #creation_time} => Time
#
# @example Request syntax with placeholder values
#
# resp = client.describe_notebook_instance_lifecycle_config({
# notebook_instance_lifecycle_config_name: "NotebookInstanceLifecycleConfigName", # required
# })
#
# @example Response structure
#
# resp.notebook_instance_lifecycle_config_arn #=> String
# resp.notebook_instance_lifecycle_config_name #=> String
# resp.on_create #=> Array
# resp.on_create[0].content #=> String
# resp.on_start #=> Array
# resp.on_start[0].content #=> String
# resp.last_modified_time #=> Time
# resp.creation_time #=> Time
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeNotebookInstanceLifecycleConfig AWS API Documentation
#
# @overload describe_notebook_instance_lifecycle_config(params = {})
# @param [Hash] params ({})
def describe_notebook_instance_lifecycle_config(params = {}, options = {})
req = build_request(:describe_notebook_instance_lifecycle_config, params)
req.send_request(options)
end
# Describes the details of a pipeline.
#
# @option params [required, String] :pipeline_name
# The name of the pipeline to describe.
#
# @return [Types::DescribePipelineResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribePipelineResponse#pipeline_arn #pipeline_arn} => String
# * {Types::DescribePipelineResponse#pipeline_name #pipeline_name} => String
# * {Types::DescribePipelineResponse#pipeline_display_name #pipeline_display_name} => String
# * {Types::DescribePipelineResponse#pipeline_definition #pipeline_definition} => String
# * {Types::DescribePipelineResponse#pipeline_description #pipeline_description} => String
# * {Types::DescribePipelineResponse#role_arn #role_arn} => String
# * {Types::DescribePipelineResponse#pipeline_status #pipeline_status} => String
# * {Types::DescribePipelineResponse#creation_time #creation_time} => Time
# * {Types::DescribePipelineResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribePipelineResponse#last_run_time #last_run_time} => Time
# * {Types::DescribePipelineResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribePipelineResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribePipelineResponse#parallelism_configuration #parallelism_configuration} => Types::ParallelismConfiguration
#
# @example Request syntax with placeholder values
#
# resp = client.describe_pipeline({
# pipeline_name: "PipelineNameOrArn", # required
# })
#
# @example Response structure
#
# resp.pipeline_arn #=> String
# resp.pipeline_name #=> String
# resp.pipeline_display_name #=> String
# resp.pipeline_definition #=> String
# resp.pipeline_description #=> String
# resp.role_arn #=> String
# resp.pipeline_status #=> String, one of "Active"
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.last_run_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.parallelism_configuration.max_parallel_execution_steps #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribePipeline AWS API Documentation
#
# @overload describe_pipeline(params = {})
# @param [Hash] params ({})
def describe_pipeline(params = {}, options = {})
req = build_request(:describe_pipeline, params)
req.send_request(options)
end
# Describes the details of an execution's pipeline definition.
#
# @option params [required, String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @return [Types::DescribePipelineDefinitionForExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribePipelineDefinitionForExecutionResponse#pipeline_definition #pipeline_definition} => String
# * {Types::DescribePipelineDefinitionForExecutionResponse#creation_time #creation_time} => Time
#
# @example Request syntax with placeholder values
#
# resp = client.describe_pipeline_definition_for_execution({
# pipeline_execution_arn: "PipelineExecutionArn", # required
# })
#
# @example Response structure
#
# resp.pipeline_definition #=> String
# resp.creation_time #=> Time
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribePipelineDefinitionForExecution AWS API Documentation
#
# @overload describe_pipeline_definition_for_execution(params = {})
# @param [Hash] params ({})
def describe_pipeline_definition_for_execution(params = {}, options = {})
req = build_request(:describe_pipeline_definition_for_execution, params)
req.send_request(options)
end
# Describes the details of a pipeline execution.
#
# @option params [required, String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @return [Types::DescribePipelineExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribePipelineExecutionResponse#pipeline_arn #pipeline_arn} => String
# * {Types::DescribePipelineExecutionResponse#pipeline_execution_arn #pipeline_execution_arn} => String
# * {Types::DescribePipelineExecutionResponse#pipeline_execution_display_name #pipeline_execution_display_name} => String
# * {Types::DescribePipelineExecutionResponse#pipeline_execution_status #pipeline_execution_status} => String
# * {Types::DescribePipelineExecutionResponse#pipeline_execution_description #pipeline_execution_description} => String
# * {Types::DescribePipelineExecutionResponse#pipeline_experiment_config #pipeline_experiment_config} => Types::PipelineExperimentConfig
# * {Types::DescribePipelineExecutionResponse#failure_reason #failure_reason} => String
# * {Types::DescribePipelineExecutionResponse#creation_time #creation_time} => Time
# * {Types::DescribePipelineExecutionResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribePipelineExecutionResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribePipelineExecutionResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribePipelineExecutionResponse#parallelism_configuration #parallelism_configuration} => Types::ParallelismConfiguration
#
# @example Request syntax with placeholder values
#
# resp = client.describe_pipeline_execution({
# pipeline_execution_arn: "PipelineExecutionArn", # required
# })
#
# @example Response structure
#
# resp.pipeline_arn #=> String
# resp.pipeline_execution_arn #=> String
# resp.pipeline_execution_display_name #=> String
# resp.pipeline_execution_status #=> String, one of "Executing", "Stopping", "Stopped", "Failed", "Succeeded"
# resp.pipeline_execution_description #=> String
# resp.pipeline_experiment_config.experiment_name #=> String
# resp.pipeline_experiment_config.trial_name #=> String
# resp.failure_reason #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.parallelism_configuration.max_parallel_execution_steps #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribePipelineExecution AWS API Documentation
#
# @overload describe_pipeline_execution(params = {})
# @param [Hash] params ({})
def describe_pipeline_execution(params = {}, options = {})
req = build_request(:describe_pipeline_execution, params)
req.send_request(options)
end
# Returns a description of a processing job.
#
# @option params [required, String] :processing_job_name
# The name of the processing job. The name must be unique within an
# Amazon Web Services Region in the Amazon Web Services account.
#
# @return [Types::DescribeProcessingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeProcessingJobResponse#processing_inputs #processing_inputs} => Array<Types::ProcessingInput>
# * {Types::DescribeProcessingJobResponse#processing_output_config #processing_output_config} => Types::ProcessingOutputConfig
# * {Types::DescribeProcessingJobResponse#processing_job_name #processing_job_name} => String
# * {Types::DescribeProcessingJobResponse#processing_resources #processing_resources} => Types::ProcessingResources
# * {Types::DescribeProcessingJobResponse#stopping_condition #stopping_condition} => Types::ProcessingStoppingCondition
# * {Types::DescribeProcessingJobResponse#app_specification #app_specification} => Types::AppSpecification
# * {Types::DescribeProcessingJobResponse#environment #environment} => Hash<String,String>
# * {Types::DescribeProcessingJobResponse#network_config #network_config} => Types::NetworkConfig
# * {Types::DescribeProcessingJobResponse#role_arn #role_arn} => String
# * {Types::DescribeProcessingJobResponse#experiment_config #experiment_config} => Types::ExperimentConfig
# * {Types::DescribeProcessingJobResponse#processing_job_arn #processing_job_arn} => String
# * {Types::DescribeProcessingJobResponse#processing_job_status #processing_job_status} => String
# * {Types::DescribeProcessingJobResponse#exit_message #exit_message} => String
# * {Types::DescribeProcessingJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeProcessingJobResponse#processing_end_time #processing_end_time} => Time
# * {Types::DescribeProcessingJobResponse#processing_start_time #processing_start_time} => Time
# * {Types::DescribeProcessingJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeProcessingJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeProcessingJobResponse#monitoring_schedule_arn #monitoring_schedule_arn} => String
# * {Types::DescribeProcessingJobResponse#auto_ml_job_arn #auto_ml_job_arn} => String
# * {Types::DescribeProcessingJobResponse#training_job_arn #training_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_processing_job({
# processing_job_name: "ProcessingJobName", # required
# })
#
# @example Response structure
#
# resp.processing_inputs #=> Array
# resp.processing_inputs[0].input_name #=> String
# resp.processing_inputs[0].app_managed #=> Boolean
# resp.processing_inputs[0].s3_input.s3_uri #=> String
# resp.processing_inputs[0].s3_input.local_path #=> String
# resp.processing_inputs[0].s3_input.s3_data_type #=> String, one of "ManifestFile", "S3Prefix"
# resp.processing_inputs[0].s3_input.s3_input_mode #=> String, one of "Pipe", "File"
# resp.processing_inputs[0].s3_input.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.processing_inputs[0].s3_input.s3_compression_type #=> String, one of "None", "Gzip"
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.catalog #=> String
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.database #=> String
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.query_string #=> String
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.work_group #=> String
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.output_s3_uri #=> String
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.kms_key_id #=> String
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.output_format #=> String, one of "PARQUET", "ORC", "AVRO", "JSON", "TEXTFILE"
# resp.processing_inputs[0].dataset_definition.athena_dataset_definition.output_compression #=> String, one of "GZIP", "SNAPPY", "ZLIB"
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.cluster_id #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.database #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.db_user #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.query_string #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.cluster_role_arn #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.output_s3_uri #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.kms_key_id #=> String
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.output_format #=> String, one of "PARQUET", "CSV"
# resp.processing_inputs[0].dataset_definition.redshift_dataset_definition.output_compression #=> String, one of "None", "GZIP", "BZIP2", "ZSTD", "SNAPPY"
# resp.processing_inputs[0].dataset_definition.local_path #=> String
# resp.processing_inputs[0].dataset_definition.data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.processing_inputs[0].dataset_definition.input_mode #=> String, one of "Pipe", "File"
# resp.processing_output_config.outputs #=> Array
# resp.processing_output_config.outputs[0].output_name #=> String
# resp.processing_output_config.outputs[0].s3_output.s3_uri #=> String
# resp.processing_output_config.outputs[0].s3_output.local_path #=> String
# resp.processing_output_config.outputs[0].s3_output.s3_upload_mode #=> String, one of "Continuous", "EndOfJob"
# resp.processing_output_config.outputs[0].feature_store_output.feature_group_name #=> String
# resp.processing_output_config.outputs[0].app_managed #=> Boolean
# resp.processing_output_config.kms_key_id #=> String
# resp.processing_job_name #=> String
# resp.processing_resources.cluster_config.instance_count #=> Integer
# resp.processing_resources.cluster_config.instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.processing_resources.cluster_config.volume_size_in_gb #=> Integer
# resp.processing_resources.cluster_config.volume_kms_key_id #=> String
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
# resp.app_specification.image_uri #=> String
# resp.app_specification.container_entrypoint #=> Array
# resp.app_specification.container_entrypoint[0] #=> String
# resp.app_specification.container_arguments #=> Array
# resp.app_specification.container_arguments[0] #=> String
# resp.environment #=> Hash
# resp.environment["ProcessingEnvironmentKey"] #=> String
# resp.network_config.enable_inter_container_traffic_encryption #=> Boolean
# resp.network_config.enable_network_isolation #=> Boolean
# resp.network_config.vpc_config.security_group_ids #=> Array
# resp.network_config.vpc_config.security_group_ids[0] #=> String
# resp.network_config.vpc_config.subnets #=> Array
# resp.network_config.vpc_config.subnets[0] #=> String
# resp.role_arn #=> String
# resp.experiment_config.experiment_name #=> String
# resp.experiment_config.trial_name #=> String
# resp.experiment_config.trial_component_display_name #=> String
# resp.experiment_config.run_name #=> String
# resp.processing_job_arn #=> String
# resp.processing_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.exit_message #=> String
# resp.failure_reason #=> String
# resp.processing_end_time #=> Time
# resp.processing_start_time #=> Time
# resp.last_modified_time #=> Time
# resp.creation_time #=> Time
# resp.monitoring_schedule_arn #=> String
# resp.auto_ml_job_arn #=> String
# resp.training_job_arn #=> String
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * processing_job_completed_or_stopped
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeProcessingJob AWS API Documentation
#
# @overload describe_processing_job(params = {})
# @param [Hash] params ({})
def describe_processing_job(params = {}, options = {})
req = build_request(:describe_processing_job, params)
req.send_request(options)
end
# Describes the details of a project.
#
# @option params [required, String] :project_name
# The name of the project to describe.
#
# @return [Types::DescribeProjectOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeProjectOutput#project_arn #project_arn} => String
# * {Types::DescribeProjectOutput#project_name #project_name} => String
# * {Types::DescribeProjectOutput#project_id #project_id} => String
# * {Types::DescribeProjectOutput#project_description #project_description} => String
# * {Types::DescribeProjectOutput#service_catalog_provisioning_details #service_catalog_provisioning_details} => Types::ServiceCatalogProvisioningDetails
# * {Types::DescribeProjectOutput#service_catalog_provisioned_product_details #service_catalog_provisioned_product_details} => Types::ServiceCatalogProvisionedProductDetails
# * {Types::DescribeProjectOutput#project_status #project_status} => String
# * {Types::DescribeProjectOutput#created_by #created_by} => Types::UserContext
# * {Types::DescribeProjectOutput#creation_time #creation_time} => Time
# * {Types::DescribeProjectOutput#last_modified_time #last_modified_time} => Time
# * {Types::DescribeProjectOutput#last_modified_by #last_modified_by} => Types::UserContext
#
# @example Request syntax with placeholder values
#
# resp = client.describe_project({
# project_name: "ProjectEntityName", # required
# })
#
# @example Response structure
#
# resp.project_arn #=> String
# resp.project_name #=> String
# resp.project_id #=> String
# resp.project_description #=> String
# resp.service_catalog_provisioning_details.product_id #=> String
# resp.service_catalog_provisioning_details.provisioning_artifact_id #=> String
# resp.service_catalog_provisioning_details.path_id #=> String
# resp.service_catalog_provisioning_details.provisioning_parameters #=> Array
# resp.service_catalog_provisioning_details.provisioning_parameters[0].key #=> String
# resp.service_catalog_provisioning_details.provisioning_parameters[0].value #=> String
# resp.service_catalog_provisioned_product_details.provisioned_product_id #=> String
# resp.service_catalog_provisioned_product_details.provisioned_product_status_message #=> String
# resp.project_status #=> String, one of "Pending", "CreateInProgress", "CreateCompleted", "CreateFailed", "DeleteInProgress", "DeleteFailed", "DeleteCompleted", "UpdateInProgress", "UpdateCompleted", "UpdateFailed"
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeProject AWS API Documentation
#
# @overload describe_project(params = {})
# @param [Hash] params ({})
def describe_project(params = {}, options = {})
req = build_request(:describe_project, params)
req.send_request(options)
end
# Describes the space.
#
# @option params [required, String] :domain_id
# The ID of the associated Domain.
#
# @option params [required, String] :space_name
# The name of the space.
#
# @return [Types::DescribeSpaceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeSpaceResponse#domain_id #domain_id} => String
# * {Types::DescribeSpaceResponse#space_arn #space_arn} => String
# * {Types::DescribeSpaceResponse#space_name #space_name} => String
# * {Types::DescribeSpaceResponse#home_efs_file_system_uid #home_efs_file_system_uid} => String
# * {Types::DescribeSpaceResponse#status #status} => String
# * {Types::DescribeSpaceResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeSpaceResponse#creation_time #creation_time} => Time
# * {Types::DescribeSpaceResponse#failure_reason #failure_reason} => String
# * {Types::DescribeSpaceResponse#space_settings #space_settings} => Types::SpaceSettings
#
# @example Request syntax with placeholder values
#
# resp = client.describe_space({
# domain_id: "DomainId", # required
# space_name: "SpaceName", # required
# })
#
# @example Response structure
#
# resp.domain_id #=> String
# resp.space_arn #=> String
# resp.space_name #=> String
# resp.home_efs_file_system_uid #=> String
# resp.status #=> String, one of "Deleting", "Failed", "InService", "Pending", "Updating", "Update_Failed", "Delete_Failed"
# resp.last_modified_time #=> Time
# resp.creation_time #=> Time
# resp.failure_reason #=> String
# resp.space_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.space_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.space_settings.jupyter_server_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.space_settings.jupyter_server_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.space_settings.jupyter_server_app_settings.lifecycle_config_arns #=> Array
# resp.space_settings.jupyter_server_app_settings.lifecycle_config_arns[0] #=> String
# resp.space_settings.jupyter_server_app_settings.code_repositories #=> Array
# resp.space_settings.jupyter_server_app_settings.code_repositories[0].repository_url #=> String
# resp.space_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.space_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.space_settings.kernel_gateway_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.space_settings.kernel_gateway_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.space_settings.kernel_gateway_app_settings.custom_images #=> Array
# resp.space_settings.kernel_gateway_app_settings.custom_images[0].image_name #=> String
# resp.space_settings.kernel_gateway_app_settings.custom_images[0].image_version_number #=> Integer
# resp.space_settings.kernel_gateway_app_settings.custom_images[0].app_image_config_name #=> String
# resp.space_settings.kernel_gateway_app_settings.lifecycle_config_arns #=> Array
# resp.space_settings.kernel_gateway_app_settings.lifecycle_config_arns[0] #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeSpace AWS API Documentation
#
# @overload describe_space(params = {})
# @param [Hash] params ({})
def describe_space(params = {}, options = {})
req = build_request(:describe_space, params)
req.send_request(options)
end
# Describes the Studio Lifecycle Configuration.
#
# @option params [required, String] :studio_lifecycle_config_name
# The name of the Studio Lifecycle Configuration to describe.
#
# @return [Types::DescribeStudioLifecycleConfigResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeStudioLifecycleConfigResponse#studio_lifecycle_config_arn #studio_lifecycle_config_arn} => String
# * {Types::DescribeStudioLifecycleConfigResponse#studio_lifecycle_config_name #studio_lifecycle_config_name} => String
# * {Types::DescribeStudioLifecycleConfigResponse#creation_time #creation_time} => Time
# * {Types::DescribeStudioLifecycleConfigResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeStudioLifecycleConfigResponse#studio_lifecycle_config_content #studio_lifecycle_config_content} => String
# * {Types::DescribeStudioLifecycleConfigResponse#studio_lifecycle_config_app_type #studio_lifecycle_config_app_type} => String
#
# @example Request syntax with placeholder values
#
# resp = client.describe_studio_lifecycle_config({
# studio_lifecycle_config_name: "StudioLifecycleConfigName", # required
# })
#
# @example Response structure
#
# resp.studio_lifecycle_config_arn #=> String
# resp.studio_lifecycle_config_name #=> String
# resp.creation_time #=> Time
# resp.last_modified_time #=> Time
# resp.studio_lifecycle_config_content #=> String
# resp.studio_lifecycle_config_app_type #=> String, one of "JupyterServer", "KernelGateway"
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeStudioLifecycleConfig AWS API Documentation
#
# @overload describe_studio_lifecycle_config(params = {})
# @param [Hash] params ({})
def describe_studio_lifecycle_config(params = {}, options = {})
req = build_request(:describe_studio_lifecycle_config, params)
req.send_request(options)
end
# Gets information about a work team provided by a vendor. It returns
# details about the subscription with a vendor in the Amazon Web
# Services Marketplace.
#
# @option params [required, String] :workteam_arn
# The Amazon Resource Name (ARN) of the subscribed work team to
# describe.
#
# @return [Types::DescribeSubscribedWorkteamResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeSubscribedWorkteamResponse#subscribed_workteam #subscribed_workteam} => Types::SubscribedWorkteam
#
# @example Request syntax with placeholder values
#
# resp = client.describe_subscribed_workteam({
# workteam_arn: "WorkteamArn", # required
# })
#
# @example Response structure
#
# resp.subscribed_workteam.workteam_arn #=> String
# resp.subscribed_workteam.marketplace_title #=> String
# resp.subscribed_workteam.seller_name #=> String
# resp.subscribed_workteam.marketplace_description #=> String
# resp.subscribed_workteam.listing_id #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeSubscribedWorkteam AWS API Documentation
#
# @overload describe_subscribed_workteam(params = {})
# @param [Hash] params ({})
def describe_subscribed_workteam(params = {}, options = {})
req = build_request(:describe_subscribed_workteam, params)
req.send_request(options)
end
# Returns information about a training job.
#
# Some of the attributes below only appear if the training job
# successfully starts. If the training job fails, `TrainingJobStatus` is
# `Failed` and, depending on the `FailureReason`, attributes like
# `TrainingStartTime`, `TrainingTimeInSeconds`, `TrainingEndTime`, and
# `BillableTimeInSeconds` may not be present in the response.
#
# @option params [required, String] :training_job_name
# The name of the training job.
#
# @return [Types::DescribeTrainingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeTrainingJobResponse#training_job_name #training_job_name} => String
# * {Types::DescribeTrainingJobResponse#training_job_arn #training_job_arn} => String
# * {Types::DescribeTrainingJobResponse#tuning_job_arn #tuning_job_arn} => String
# * {Types::DescribeTrainingJobResponse#labeling_job_arn #labeling_job_arn} => String
# * {Types::DescribeTrainingJobResponse#auto_ml_job_arn #auto_ml_job_arn} => String
# * {Types::DescribeTrainingJobResponse#model_artifacts #model_artifacts} => Types::ModelArtifacts
# * {Types::DescribeTrainingJobResponse#training_job_status #training_job_status} => String
# * {Types::DescribeTrainingJobResponse#secondary_status #secondary_status} => String
# * {Types::DescribeTrainingJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeTrainingJobResponse#hyper_parameters #hyper_parameters} => Hash<String,String>
# * {Types::DescribeTrainingJobResponse#algorithm_specification #algorithm_specification} => Types::AlgorithmSpecification
# * {Types::DescribeTrainingJobResponse#role_arn #role_arn} => String
# * {Types::DescribeTrainingJobResponse#input_data_config #input_data_config} => Array<Types::Channel>
# * {Types::DescribeTrainingJobResponse#output_data_config #output_data_config} => Types::OutputDataConfig
# * {Types::DescribeTrainingJobResponse#resource_config #resource_config} => Types::ResourceConfig
# * {Types::DescribeTrainingJobResponse#vpc_config #vpc_config} => Types::VpcConfig
# * {Types::DescribeTrainingJobResponse#stopping_condition #stopping_condition} => Types::StoppingCondition
# * {Types::DescribeTrainingJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeTrainingJobResponse#training_start_time #training_start_time} => Time
# * {Types::DescribeTrainingJobResponse#training_end_time #training_end_time} => Time
# * {Types::DescribeTrainingJobResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeTrainingJobResponse#secondary_status_transitions #secondary_status_transitions} => Array<Types::SecondaryStatusTransition>
# * {Types::DescribeTrainingJobResponse#final_metric_data_list #final_metric_data_list} => Array<Types::MetricData>
# * {Types::DescribeTrainingJobResponse#enable_network_isolation #enable_network_isolation} => Boolean
# * {Types::DescribeTrainingJobResponse#enable_inter_container_traffic_encryption #enable_inter_container_traffic_encryption} => Boolean
# * {Types::DescribeTrainingJobResponse#enable_managed_spot_training #enable_managed_spot_training} => Boolean
# * {Types::DescribeTrainingJobResponse#checkpoint_config #checkpoint_config} => Types::CheckpointConfig
# * {Types::DescribeTrainingJobResponse#training_time_in_seconds #training_time_in_seconds} => Integer
# * {Types::DescribeTrainingJobResponse#billable_time_in_seconds #billable_time_in_seconds} => Integer
# * {Types::DescribeTrainingJobResponse#debug_hook_config #debug_hook_config} => Types::DebugHookConfig
# * {Types::DescribeTrainingJobResponse#experiment_config #experiment_config} => Types::ExperimentConfig
# * {Types::DescribeTrainingJobResponse#debug_rule_configurations #debug_rule_configurations} => Array<Types::DebugRuleConfiguration>
# * {Types::DescribeTrainingJobResponse#tensor_board_output_config #tensor_board_output_config} => Types::TensorBoardOutputConfig
# * {Types::DescribeTrainingJobResponse#debug_rule_evaluation_statuses #debug_rule_evaluation_statuses} => Array<Types::DebugRuleEvaluationStatus>
# * {Types::DescribeTrainingJobResponse#profiler_config #profiler_config} => Types::ProfilerConfig
# * {Types::DescribeTrainingJobResponse#profiler_rule_configurations #profiler_rule_configurations} => Array<Types::ProfilerRuleConfiguration>
# * {Types::DescribeTrainingJobResponse#profiler_rule_evaluation_statuses #profiler_rule_evaluation_statuses} => Array<Types::ProfilerRuleEvaluationStatus>
# * {Types::DescribeTrainingJobResponse#profiling_status #profiling_status} => String
# * {Types::DescribeTrainingJobResponse#retry_strategy #retry_strategy} => Types::RetryStrategy
# * {Types::DescribeTrainingJobResponse#environment #environment} => Hash<String,String>
# * {Types::DescribeTrainingJobResponse#warm_pool_status #warm_pool_status} => Types::WarmPoolStatus
#
# @example Request syntax with placeholder values
#
# resp = client.describe_training_job({
# training_job_name: "TrainingJobName", # required
# })
#
# @example Response structure
#
# resp.training_job_name #=> String
# resp.training_job_arn #=> String
# resp.tuning_job_arn #=> String
# resp.labeling_job_arn #=> String
# resp.auto_ml_job_arn #=> String
# resp.model_artifacts.s3_model_artifacts #=> String
# resp.training_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.secondary_status #=> String, one of "Starting", "LaunchingMLInstances", "PreparingTrainingStack", "Downloading", "DownloadingTrainingImage", "Training", "Uploading", "Stopping", "Stopped", "MaxRuntimeExceeded", "Completed", "Failed", "Interrupted", "MaxWaitTimeExceeded", "Updating", "Restarting"
# resp.failure_reason #=> String
# resp.hyper_parameters #=> Hash
# resp.hyper_parameters["HyperParameterKey"] #=> String
# resp.algorithm_specification.training_image #=> String
# resp.algorithm_specification.algorithm_name #=> String
# resp.algorithm_specification.training_input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.algorithm_specification.metric_definitions #=> Array
# resp.algorithm_specification.metric_definitions[0].name #=> String
# resp.algorithm_specification.metric_definitions[0].regex #=> String
# resp.algorithm_specification.enable_sage_maker_metrics_time_series #=> Boolean
# resp.algorithm_specification.container_entrypoint #=> Array
# resp.algorithm_specification.container_entrypoint[0] #=> String
# resp.algorithm_specification.container_arguments #=> Array
# resp.algorithm_specification.container_arguments[0] #=> String
# resp.algorithm_specification.training_image_config.training_repository_access_mode #=> String, one of "Platform", "Vpc"
# resp.algorithm_specification.training_image_config.training_repository_auth_config.training_repository_credentials_provider_arn #=> String
# resp.role_arn #=> String
# resp.input_data_config #=> Array
# resp.input_data_config[0].channel_name #=> String
# resp.input_data_config[0].data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.input_data_config[0].data_source.s3_data_source.s3_uri #=> String
# resp.input_data_config[0].data_source.s3_data_source.s3_data_distribution_type #=> String, one of "FullyReplicated", "ShardedByS3Key"
# resp.input_data_config[0].data_source.s3_data_source.attribute_names #=> Array
# resp.input_data_config[0].data_source.s3_data_source.attribute_names[0] #=> String
# resp.input_data_config[0].data_source.s3_data_source.instance_group_names #=> Array
# resp.input_data_config[0].data_source.s3_data_source.instance_group_names[0] #=> String
# resp.input_data_config[0].data_source.file_system_data_source.file_system_id #=> String
# resp.input_data_config[0].data_source.file_system_data_source.file_system_access_mode #=> String, one of "rw", "ro"
# resp.input_data_config[0].data_source.file_system_data_source.file_system_type #=> String, one of "EFS", "FSxLustre"
# resp.input_data_config[0].data_source.file_system_data_source.directory_path #=> String
# resp.input_data_config[0].content_type #=> String
# resp.input_data_config[0].compression_type #=> String, one of "None", "Gzip"
# resp.input_data_config[0].record_wrapper_type #=> String, one of "None", "RecordIO"
# resp.input_data_config[0].input_mode #=> String, one of "Pipe", "File", "FastFile"
# resp.input_data_config[0].shuffle_config.seed #=> Integer
# resp.output_data_config.kms_key_id #=> String
# resp.output_data_config.s3_output_path #=> String
# resp.resource_config.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.resource_config.instance_count #=> Integer
# resp.resource_config.volume_size_in_gb #=> Integer
# resp.resource_config.volume_kms_key_id #=> String
# resp.resource_config.instance_groups #=> Array
# resp.resource_config.instance_groups[0].instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.p4d.24xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5n.xlarge", "ml.c5n.2xlarge", "ml.c5n.4xlarge", "ml.c5n.9xlarge", "ml.c5n.18xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge"
# resp.resource_config.instance_groups[0].instance_count #=> Integer
# resp.resource_config.instance_groups[0].instance_group_name #=> String
# resp.resource_config.keep_alive_period_in_seconds #=> Integer
# resp.vpc_config.security_group_ids #=> Array
# resp.vpc_config.security_group_ids[0] #=> String
# resp.vpc_config.subnets #=> Array
# resp.vpc_config.subnets[0] #=> String
# resp.stopping_condition.max_runtime_in_seconds #=> Integer
# resp.stopping_condition.max_wait_time_in_seconds #=> Integer
# resp.creation_time #=> Time
# resp.training_start_time #=> Time
# resp.training_end_time #=> Time
# resp.last_modified_time #=> Time
# resp.secondary_status_transitions #=> Array
# resp.secondary_status_transitions[0].status #=> String, one of "Starting", "LaunchingMLInstances", "PreparingTrainingStack", "Downloading", "DownloadingTrainingImage", "Training", "Uploading", "Stopping", "Stopped", "MaxRuntimeExceeded", "Completed", "Failed", "Interrupted", "MaxWaitTimeExceeded", "Updating", "Restarting"
# resp.secondary_status_transitions[0].start_time #=> Time
# resp.secondary_status_transitions[0].end_time #=> Time
# resp.secondary_status_transitions[0].status_message #=> String
# resp.final_metric_data_list #=> Array
# resp.final_metric_data_list[0].metric_name #=> String
# resp.final_metric_data_list[0].value #=> Float
# resp.final_metric_data_list[0].timestamp #=> Time
# resp.enable_network_isolation #=> Boolean
# resp.enable_inter_container_traffic_encryption #=> Boolean
# resp.enable_managed_spot_training #=> Boolean
# resp.checkpoint_config.s3_uri #=> String
# resp.checkpoint_config.local_path #=> String
# resp.training_time_in_seconds #=> Integer
# resp.billable_time_in_seconds #=> Integer
# resp.debug_hook_config.local_path #=> String
# resp.debug_hook_config.s3_output_path #=> String
# resp.debug_hook_config.hook_parameters #=> Hash
# resp.debug_hook_config.hook_parameters["ConfigKey"] #=> String
# resp.debug_hook_config.collection_configurations #=> Array
# resp.debug_hook_config.collection_configurations[0].collection_name #=> String
# resp.debug_hook_config.collection_configurations[0].collection_parameters #=> Hash
# resp.debug_hook_config.collection_configurations[0].collection_parameters["ConfigKey"] #=> String
# resp.experiment_config.experiment_name #=> String
# resp.experiment_config.trial_name #=> String
# resp.experiment_config.trial_component_display_name #=> String
# resp.experiment_config.run_name #=> String
# resp.debug_rule_configurations #=> Array
# resp.debug_rule_configurations[0].rule_configuration_name #=> String
# resp.debug_rule_configurations[0].local_path #=> String
# resp.debug_rule_configurations[0].s3_output_path #=> String
# resp.debug_rule_configurations[0].rule_evaluator_image #=> String
# resp.debug_rule_configurations[0].instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.debug_rule_configurations[0].volume_size_in_gb #=> Integer
# resp.debug_rule_configurations[0].rule_parameters #=> Hash
# resp.debug_rule_configurations[0].rule_parameters["ConfigKey"] #=> String
# resp.tensor_board_output_config.local_path #=> String
# resp.tensor_board_output_config.s3_output_path #=> String
# resp.debug_rule_evaluation_statuses #=> Array
# resp.debug_rule_evaluation_statuses[0].rule_configuration_name #=> String
# resp.debug_rule_evaluation_statuses[0].rule_evaluation_job_arn #=> String
# resp.debug_rule_evaluation_statuses[0].rule_evaluation_status #=> String, one of "InProgress", "NoIssuesFound", "IssuesFound", "Error", "Stopping", "Stopped"
# resp.debug_rule_evaluation_statuses[0].status_details #=> String
# resp.debug_rule_evaluation_statuses[0].last_modified_time #=> Time
# resp.profiler_config.s3_output_path #=> String
# resp.profiler_config.profiling_interval_in_milliseconds #=> Integer
# resp.profiler_config.profiling_parameters #=> Hash
# resp.profiler_config.profiling_parameters["ConfigKey"] #=> String
# resp.profiler_config.disable_profiler #=> Boolean
# resp.profiler_rule_configurations #=> Array
# resp.profiler_rule_configurations[0].rule_configuration_name #=> String
# resp.profiler_rule_configurations[0].local_path #=> String
# resp.profiler_rule_configurations[0].s3_output_path #=> String
# resp.profiler_rule_configurations[0].rule_evaluator_image #=> String
# resp.profiler_rule_configurations[0].instance_type #=> String, one of "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.profiler_rule_configurations[0].volume_size_in_gb #=> Integer
# resp.profiler_rule_configurations[0].rule_parameters #=> Hash
# resp.profiler_rule_configurations[0].rule_parameters["ConfigKey"] #=> String
# resp.profiler_rule_evaluation_statuses #=> Array
# resp.profiler_rule_evaluation_statuses[0].rule_configuration_name #=> String
# resp.profiler_rule_evaluation_statuses[0].rule_evaluation_job_arn #=> String
# resp.profiler_rule_evaluation_statuses[0].rule_evaluation_status #=> String, one of "InProgress", "NoIssuesFound", "IssuesFound", "Error", "Stopping", "Stopped"
# resp.profiler_rule_evaluation_statuses[0].status_details #=> String
# resp.profiler_rule_evaluation_statuses[0].last_modified_time #=> Time
# resp.profiling_status #=> String, one of "Enabled", "Disabled"
# resp.retry_strategy.maximum_retry_attempts #=> Integer
# resp.environment #=> Hash
# resp.environment["TrainingEnvironmentKey"] #=> String
# resp.warm_pool_status.status #=> String, one of "Available", "Terminated", "Reused", "InUse"
# resp.warm_pool_status.resource_retained_billable_time_in_seconds #=> Integer
# resp.warm_pool_status.reused_by_job #=> String
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * training_job_completed_or_stopped
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeTrainingJob AWS API Documentation
#
# @overload describe_training_job(params = {})
# @param [Hash] params ({})
def describe_training_job(params = {}, options = {})
req = build_request(:describe_training_job, params)
req.send_request(options)
end
# Returns information about a transform job.
#
# @option params [required, String] :transform_job_name
# The name of the transform job that you want to view details of.
#
# @return [Types::DescribeTransformJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeTransformJobResponse#transform_job_name #transform_job_name} => String
# * {Types::DescribeTransformJobResponse#transform_job_arn #transform_job_arn} => String
# * {Types::DescribeTransformJobResponse#transform_job_status #transform_job_status} => String
# * {Types::DescribeTransformJobResponse#failure_reason #failure_reason} => String
# * {Types::DescribeTransformJobResponse#model_name #model_name} => String
# * {Types::DescribeTransformJobResponse#max_concurrent_transforms #max_concurrent_transforms} => Integer
# * {Types::DescribeTransformJobResponse#model_client_config #model_client_config} => Types::ModelClientConfig
# * {Types::DescribeTransformJobResponse#max_payload_in_mb #max_payload_in_mb} => Integer
# * {Types::DescribeTransformJobResponse#batch_strategy #batch_strategy} => String
# * {Types::DescribeTransformJobResponse#environment #environment} => Hash<String,String>
# * {Types::DescribeTransformJobResponse#transform_input #transform_input} => Types::TransformInput
# * {Types::DescribeTransformJobResponse#transform_output #transform_output} => Types::TransformOutput
# * {Types::DescribeTransformJobResponse#data_capture_config #data_capture_config} => Types::BatchDataCaptureConfig
# * {Types::DescribeTransformJobResponse#transform_resources #transform_resources} => Types::TransformResources
# * {Types::DescribeTransformJobResponse#creation_time #creation_time} => Time
# * {Types::DescribeTransformJobResponse#transform_start_time #transform_start_time} => Time
# * {Types::DescribeTransformJobResponse#transform_end_time #transform_end_time} => Time
# * {Types::DescribeTransformJobResponse#labeling_job_arn #labeling_job_arn} => String
# * {Types::DescribeTransformJobResponse#auto_ml_job_arn #auto_ml_job_arn} => String
# * {Types::DescribeTransformJobResponse#data_processing #data_processing} => Types::DataProcessing
# * {Types::DescribeTransformJobResponse#experiment_config #experiment_config} => Types::ExperimentConfig
#
# @example Request syntax with placeholder values
#
# resp = client.describe_transform_job({
# transform_job_name: "TransformJobName", # required
# })
#
# @example Response structure
#
# resp.transform_job_name #=> String
# resp.transform_job_arn #=> String
# resp.transform_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.failure_reason #=> String
# resp.model_name #=> String
# resp.max_concurrent_transforms #=> Integer
# resp.model_client_config.invocations_timeout_in_seconds #=> Integer
# resp.model_client_config.invocations_max_retries #=> Integer
# resp.max_payload_in_mb #=> Integer
# resp.batch_strategy #=> String, one of "MultiRecord", "SingleRecord"
# resp.environment #=> Hash
# resp.environment["TransformEnvironmentKey"] #=> String
# resp.transform_input.data_source.s3_data_source.s3_data_type #=> String, one of "ManifestFile", "S3Prefix", "AugmentedManifestFile"
# resp.transform_input.data_source.s3_data_source.s3_uri #=> String
# resp.transform_input.content_type #=> String
# resp.transform_input.compression_type #=> String, one of "None", "Gzip"
# resp.transform_input.split_type #=> String, one of "None", "Line", "RecordIO", "TFRecord"
# resp.transform_output.s3_output_path #=> String
# resp.transform_output.accept #=> String
# resp.transform_output.assemble_with #=> String, one of "None", "Line"
# resp.transform_output.kms_key_id #=> String
# resp.data_capture_config.destination_s3_uri #=> String
# resp.data_capture_config.kms_key_id #=> String
# resp.data_capture_config.generate_inference_id #=> Boolean
# resp.transform_resources.instance_type #=> String, one of "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge"
# resp.transform_resources.instance_count #=> Integer
# resp.transform_resources.volume_kms_key_id #=> String
# resp.creation_time #=> Time
# resp.transform_start_time #=> Time
# resp.transform_end_time #=> Time
# resp.labeling_job_arn #=> String
# resp.auto_ml_job_arn #=> String
# resp.data_processing.input_filter #=> String
# resp.data_processing.output_filter #=> String
# resp.data_processing.join_source #=> String, one of "Input", "None"
# resp.experiment_config.experiment_name #=> String
# resp.experiment_config.trial_name #=> String
# resp.experiment_config.trial_component_display_name #=> String
# resp.experiment_config.run_name #=> String
#
#
# The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
#
# * transform_job_completed_or_stopped
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeTransformJob AWS API Documentation
#
# @overload describe_transform_job(params = {})
# @param [Hash] params ({})
def describe_transform_job(params = {}, options = {})
req = build_request(:describe_transform_job, params)
req.send_request(options)
end
# Provides a list of a trial's properties.
#
# @option params [required, String] :trial_name
# The name of the trial to describe.
#
# @return [Types::DescribeTrialResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeTrialResponse#trial_name #trial_name} => String
# * {Types::DescribeTrialResponse#trial_arn #trial_arn} => String
# * {Types::DescribeTrialResponse#display_name #display_name} => String
# * {Types::DescribeTrialResponse#experiment_name #experiment_name} => String
# * {Types::DescribeTrialResponse#source #source} => Types::TrialSource
# * {Types::DescribeTrialResponse#creation_time #creation_time} => Time
# * {Types::DescribeTrialResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeTrialResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeTrialResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeTrialResponse#metadata_properties #metadata_properties} => Types::MetadataProperties
#
# @example Request syntax with placeholder values
#
# resp = client.describe_trial({
# trial_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.trial_name #=> String
# resp.trial_arn #=> String
# resp.display_name #=> String
# resp.experiment_name #=> String
# resp.source.source_arn #=> String
# resp.source.source_type #=> String
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.metadata_properties.commit_id #=> String
# resp.metadata_properties.repository #=> String
# resp.metadata_properties.generated_by #=> String
# resp.metadata_properties.project_id #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeTrial AWS API Documentation
#
# @overload describe_trial(params = {})
# @param [Hash] params ({})
def describe_trial(params = {}, options = {})
req = build_request(:describe_trial, params)
req.send_request(options)
end
# Provides a list of a trials component's properties.
#
# @option params [required, String] :trial_component_name
# The name of the trial component to describe.
#
# @return [Types::DescribeTrialComponentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeTrialComponentResponse#trial_component_name #trial_component_name} => String
# * {Types::DescribeTrialComponentResponse#trial_component_arn #trial_component_arn} => String
# * {Types::DescribeTrialComponentResponse#display_name #display_name} => String
# * {Types::DescribeTrialComponentResponse#source #source} => Types::TrialComponentSource
# * {Types::DescribeTrialComponentResponse#status #status} => Types::TrialComponentStatus
# * {Types::DescribeTrialComponentResponse#start_time #start_time} => Time
# * {Types::DescribeTrialComponentResponse#end_time #end_time} => Time
# * {Types::DescribeTrialComponentResponse#creation_time #creation_time} => Time
# * {Types::DescribeTrialComponentResponse#created_by #created_by} => Types::UserContext
# * {Types::DescribeTrialComponentResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeTrialComponentResponse#last_modified_by #last_modified_by} => Types::UserContext
# * {Types::DescribeTrialComponentResponse#parameters #parameters} => Hash<String,Types::TrialComponentParameterValue>
# * {Types::DescribeTrialComponentResponse#input_artifacts #input_artifacts} => Hash<String,Types::TrialComponentArtifact>
# * {Types::DescribeTrialComponentResponse#output_artifacts #output_artifacts} => Hash<String,Types::TrialComponentArtifact>
# * {Types::DescribeTrialComponentResponse#metadata_properties #metadata_properties} => Types::MetadataProperties
# * {Types::DescribeTrialComponentResponse#metrics #metrics} => Array<Types::TrialComponentMetricSummary>
# * {Types::DescribeTrialComponentResponse#lineage_group_arn #lineage_group_arn} => String
# * {Types::DescribeTrialComponentResponse#sources #sources} => Array<Types::TrialComponentSource>
#
# @example Request syntax with placeholder values
#
# resp = client.describe_trial_component({
# trial_component_name: "ExperimentEntityNameOrArn", # required
# })
#
# @example Response structure
#
# resp.trial_component_name #=> String
# resp.trial_component_arn #=> String
# resp.display_name #=> String
# resp.source.source_arn #=> String
# resp.source.source_type #=> String
# resp.status.primary_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.status.message #=> String
# resp.start_time #=> Time
# resp.end_time #=> Time
# resp.creation_time #=> Time
# resp.created_by.user_profile_arn #=> String
# resp.created_by.user_profile_name #=> String
# resp.created_by.domain_id #=> String
# resp.created_by.iam_identity.arn #=> String
# resp.created_by.iam_identity.principal_id #=> String
# resp.created_by.iam_identity.source_identity #=> String
# resp.last_modified_time #=> Time
# resp.last_modified_by.user_profile_arn #=> String
# resp.last_modified_by.user_profile_name #=> String
# resp.last_modified_by.domain_id #=> String
# resp.last_modified_by.iam_identity.arn #=> String
# resp.last_modified_by.iam_identity.principal_id #=> String
# resp.last_modified_by.iam_identity.source_identity #=> String
# resp.parameters #=> Hash
# resp.parameters["TrialComponentKey256"].string_value #=> String
# resp.parameters["TrialComponentKey256"].number_value #=> Float
# resp.input_artifacts #=> Hash
# resp.input_artifacts["TrialComponentKey64"].media_type #=> String
# resp.input_artifacts["TrialComponentKey64"].value #=> String
# resp.output_artifacts #=> Hash
# resp.output_artifacts["TrialComponentKey64"].media_type #=> String
# resp.output_artifacts["TrialComponentKey64"].value #=> String
# resp.metadata_properties.commit_id #=> String
# resp.metadata_properties.repository #=> String
# resp.metadata_properties.generated_by #=> String
# resp.metadata_properties.project_id #=> String
# resp.metrics #=> Array
# resp.metrics[0].metric_name #=> String
# resp.metrics[0].source_arn #=> String
# resp.metrics[0].time_stamp #=> Time
# resp.metrics[0].max #=> Float
# resp.metrics[0].min #=> Float
# resp.metrics[0].last #=> Float
# resp.metrics[0].count #=> Integer
# resp.metrics[0].avg #=> Float
# resp.metrics[0].std_dev #=> Float
# resp.lineage_group_arn #=> String
# resp.sources #=> Array
# resp.sources[0].source_arn #=> String
# resp.sources[0].source_type #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeTrialComponent AWS API Documentation
#
# @overload describe_trial_component(params = {})
# @param [Hash] params ({})
def describe_trial_component(params = {}, options = {})
req = build_request(:describe_trial_component, params)
req.send_request(options)
end
# Describes a user profile. For more information, see
# `CreateUserProfile`.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [required, String] :user_profile_name
# The user profile name. This value is not case sensitive.
#
# @return [Types::DescribeUserProfileResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeUserProfileResponse#domain_id #domain_id} => String
# * {Types::DescribeUserProfileResponse#user_profile_arn #user_profile_arn} => String
# * {Types::DescribeUserProfileResponse#user_profile_name #user_profile_name} => String
# * {Types::DescribeUserProfileResponse#home_efs_file_system_uid #home_efs_file_system_uid} => String
# * {Types::DescribeUserProfileResponse#status #status} => String
# * {Types::DescribeUserProfileResponse#last_modified_time #last_modified_time} => Time
# * {Types::DescribeUserProfileResponse#creation_time #creation_time} => Time
# * {Types::DescribeUserProfileResponse#failure_reason #failure_reason} => String
# * {Types::DescribeUserProfileResponse#single_sign_on_user_identifier #single_sign_on_user_identifier} => String
# * {Types::DescribeUserProfileResponse#single_sign_on_user_value #single_sign_on_user_value} => String
# * {Types::DescribeUserProfileResponse#user_settings #user_settings} => Types::UserSettings
#
# @example Request syntax with placeholder values
#
# resp = client.describe_user_profile({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName", # required
# })
#
# @example Response structure
#
# resp.domain_id #=> String
# resp.user_profile_arn #=> String
# resp.user_profile_name #=> String
# resp.home_efs_file_system_uid #=> String
# resp.status #=> String, one of "Deleting", "Failed", "InService", "Pending", "Updating", "Update_Failed", "Delete_Failed"
# resp.last_modified_time #=> Time
# resp.creation_time #=> Time
# resp.failure_reason #=> String
# resp.single_sign_on_user_identifier #=> String
# resp.single_sign_on_user_value #=> String
# resp.user_settings.execution_role #=> String
# resp.user_settings.security_groups #=> Array
# resp.user_settings.security_groups[0] #=> String
# resp.user_settings.sharing_settings.notebook_output_option #=> String, one of "Allowed", "Disabled"
# resp.user_settings.sharing_settings.s3_output_path #=> String
# resp.user_settings.sharing_settings.s3_kms_key_id #=> String
# resp.user_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.user_settings.jupyter_server_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.user_settings.jupyter_server_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.user_settings.jupyter_server_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.user_settings.jupyter_server_app_settings.lifecycle_config_arns #=> Array
# resp.user_settings.jupyter_server_app_settings.lifecycle_config_arns[0] #=> String
# resp.user_settings.jupyter_server_app_settings.code_repositories #=> Array
# resp.user_settings.jupyter_server_app_settings.code_repositories[0].repository_url #=> String
# resp.user_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.user_settings.kernel_gateway_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.user_settings.kernel_gateway_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.user_settings.kernel_gateway_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.user_settings.kernel_gateway_app_settings.custom_images #=> Array
# resp.user_settings.kernel_gateway_app_settings.custom_images[0].image_name #=> String
# resp.user_settings.kernel_gateway_app_settings.custom_images[0].image_version_number #=> Integer
# resp.user_settings.kernel_gateway_app_settings.custom_images[0].app_image_config_name #=> String
# resp.user_settings.kernel_gateway_app_settings.lifecycle_config_arns #=> Array
# resp.user_settings.kernel_gateway_app_settings.lifecycle_config_arns[0] #=> String
# resp.user_settings.tensor_board_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.user_settings.tensor_board_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.user_settings.tensor_board_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.user_settings.tensor_board_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.user_settings.r_studio_server_pro_app_settings.access_status #=> String, one of "ENABLED", "DISABLED"
# resp.user_settings.r_studio_server_pro_app_settings.user_group #=> String, one of "R_STUDIO_ADMIN", "R_STUDIO_USER"
# resp.user_settings.r_session_app_settings.default_resource_spec.sage_maker_image_arn #=> String
# resp.user_settings.r_session_app_settings.default_resource_spec.sage_maker_image_version_arn #=> String
# resp.user_settings.r_session_app_settings.default_resource_spec.instance_type #=> String, one of "system", "ml.t3.micro", "ml.t3.small", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.8xlarge", "ml.m5.12xlarge", "ml.m5.16xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.12xlarge", "ml.c5.18xlarge", "ml.c5.24xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.geospatial.interactive", "ml.p4d.24xlarge", "ml.p4de.24xlarge"
# resp.user_settings.r_session_app_settings.default_resource_spec.lifecycle_config_arn #=> String
# resp.user_settings.r_session_app_settings.custom_images #=> Array
# resp.user_settings.r_session_app_settings.custom_images[0].image_name #=> String
# resp.user_settings.r_session_app_settings.custom_images[0].image_version_number #=> Integer
# resp.user_settings.r_session_app_settings.custom_images[0].app_image_config_name #=> String
# resp.user_settings.canvas_app_settings.time_series_forecasting_settings.status #=> String, one of "ENABLED", "DISABLED"
# resp.user_settings.canvas_app_settings.time_series_forecasting_settings.amazon_forecast_role_arn #=> String
# resp.user_settings.canvas_app_settings.model_register_settings.status #=> String, one of "ENABLED", "DISABLED"
# resp.user_settings.canvas_app_settings.model_register_settings.cross_account_model_register_role_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeUserProfile AWS API Documentation
#
# @overload describe_user_profile(params = {})
# @param [Hash] params ({})
def describe_user_profile(params = {}, options = {})
req = build_request(:describe_user_profile, params)
req.send_request(options)
end
# Lists private workforce information, including workforce name, Amazon
# Resource Name (ARN), and, if applicable, allowed IP address ranges
# ([CIDRs][1]). Allowable IP address ranges are the IP addresses that
# workers can use to access tasks.
#
# This operation applies only to private workforces.
#
#
#
# [1]: https://docs.aws.amazon.com/vpc/latest/userguide/VPC_Subnets.html
#
# @option params [required, String] :workforce_name
# The name of the private workforce whose access you want to restrict.
# `WorkforceName` is automatically set to `default` when a workforce is
# created and cannot be modified.
#
# @return [Types::DescribeWorkforceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeWorkforceResponse#workforce #workforce} => Types::Workforce
#
# @example Request syntax with placeholder values
#
# resp = client.describe_workforce({
# workforce_name: "WorkforceName", # required
# })
#
# @example Response structure
#
# resp.workforce.workforce_name #=> String
# resp.workforce.workforce_arn #=> String
# resp.workforce.last_updated_date #=> Time
# resp.workforce.source_ip_config.cidrs #=> Array
# resp.workforce.source_ip_config.cidrs[0] #=> String
# resp.workforce.sub_domain #=> String
# resp.workforce.cognito_config.user_pool #=> String
# resp.workforce.cognito_config.client_id #=> String
# resp.workforce.oidc_config.client_id #=> String
# resp.workforce.oidc_config.issuer #=> String
# resp.workforce.oidc_config.authorization_endpoint #=> String
# resp.workforce.oidc_config.token_endpoint #=> String
# resp.workforce.oidc_config.user_info_endpoint #=> String
# resp.workforce.oidc_config.logout_endpoint #=> String
# resp.workforce.oidc_config.jwks_uri #=> String
# resp.workforce.create_date #=> Time
# resp.workforce.workforce_vpc_config.vpc_id #=> String
# resp.workforce.workforce_vpc_config.security_group_ids #=> Array
# resp.workforce.workforce_vpc_config.security_group_ids[0] #=> String
# resp.workforce.workforce_vpc_config.subnets #=> Array
# resp.workforce.workforce_vpc_config.subnets[0] #=> String
# resp.workforce.workforce_vpc_config.vpc_endpoint_id #=> String
# resp.workforce.status #=> String, one of "Initializing", "Updating", "Deleting", "Failed", "Active"
# resp.workforce.failure_reason #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeWorkforce AWS API Documentation
#
# @overload describe_workforce(params = {})
# @param [Hash] params ({})
def describe_workforce(params = {}, options = {})
req = build_request(:describe_workforce, params)
req.send_request(options)
end
# Gets information about a specific work team. You can see information
# such as the create date, the last updated date, membership
# information, and the work team's Amazon Resource Name (ARN).
#
# @option params [required, String] :workteam_name
# The name of the work team to return a description of.
#
# @return [Types::DescribeWorkteamResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DescribeWorkteamResponse#workteam #workteam} => Types::Workteam
#
# @example Request syntax with placeholder values
#
# resp = client.describe_workteam({
# workteam_name: "WorkteamName", # required
# })
#
# @example Response structure
#
# resp.workteam.workteam_name #=> String
# resp.workteam.member_definitions #=> Array
# resp.workteam.member_definitions[0].cognito_member_definition.user_pool #=> String
# resp.workteam.member_definitions[0].cognito_member_definition.user_group #=> String
# resp.workteam.member_definitions[0].cognito_member_definition.client_id #=> String
# resp.workteam.member_definitions[0].oidc_member_definition.groups #=> Array
# resp.workteam.member_definitions[0].oidc_member_definition.groups[0] #=> String
# resp.workteam.workteam_arn #=> String
# resp.workteam.workforce_arn #=> String
# resp.workteam.product_listing_ids #=> Array
# resp.workteam.product_listing_ids[0] #=> String
# resp.workteam.description #=> String
# resp.workteam.sub_domain #=> String
# resp.workteam.create_date #=> Time
# resp.workteam.last_updated_date #=> Time
# resp.workteam.notification_configuration.notification_topic_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DescribeWorkteam AWS API Documentation
#
# @overload describe_workteam(params = {})
# @param [Hash] params ({})
def describe_workteam(params = {}, options = {})
req = build_request(:describe_workteam, params)
req.send_request(options)
end
# Disables using Service Catalog in SageMaker. Service Catalog is used
# to create SageMaker projects.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DisableSagemakerServicecatalogPortfolio AWS API Documentation
#
# @overload disable_sagemaker_servicecatalog_portfolio(params = {})
# @param [Hash] params ({})
def disable_sagemaker_servicecatalog_portfolio(params = {}, options = {})
req = build_request(:disable_sagemaker_servicecatalog_portfolio, params)
req.send_request(options)
end
# Disassociates a trial component from a trial. This doesn't effect
# other trials the component is associated with. Before you can delete a
# component, you must disassociate the component from all trials it is
# associated with. To associate a trial component with a trial, call the
# [AssociateTrialComponent][1] API.
#
# To get a list of the trials a component is associated with, use the
# [Search][2] API. Specify `ExperimentTrialComponent` for the `Resource`
# parameter. The list appears in the response under
# `Results.TrialComponent.Parents`.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_AssociateTrialComponent.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_Search.html
#
# @option params [required, String] :trial_component_name
# The name of the component to disassociate from the trial.
#
# @option params [required, String] :trial_name
# The name of the trial to disassociate from.
#
# @return [Types::DisassociateTrialComponentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::DisassociateTrialComponentResponse#trial_component_arn #trial_component_arn} => String
# * {Types::DisassociateTrialComponentResponse#trial_arn #trial_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.disassociate_trial_component({
# trial_component_name: "ExperimentEntityName", # required
# trial_name: "ExperimentEntityName", # required
# })
#
# @example Response structure
#
# resp.trial_component_arn #=> String
# resp.trial_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/DisassociateTrialComponent AWS API Documentation
#
# @overload disassociate_trial_component(params = {})
# @param [Hash] params ({})
def disassociate_trial_component(params = {}, options = {})
req = build_request(:disassociate_trial_component, params)
req.send_request(options)
end
# Enables using Service Catalog in SageMaker. Service Catalog is used to
# create SageMaker projects.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/EnableSagemakerServicecatalogPortfolio AWS API Documentation
#
# @overload enable_sagemaker_servicecatalog_portfolio(params = {})
# @param [Hash] params ({})
def enable_sagemaker_servicecatalog_portfolio(params = {}, options = {})
req = build_request(:enable_sagemaker_servicecatalog_portfolio, params)
req.send_request(options)
end
# Describes a fleet.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet.
#
# @return [Types::GetDeviceFleetReportResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::GetDeviceFleetReportResponse#device_fleet_arn #device_fleet_arn} => String
# * {Types::GetDeviceFleetReportResponse#device_fleet_name #device_fleet_name} => String
# * {Types::GetDeviceFleetReportResponse#output_config #output_config} => Types::EdgeOutputConfig
# * {Types::GetDeviceFleetReportResponse#description #description} => String
# * {Types::GetDeviceFleetReportResponse#report_generated #report_generated} => Time
# * {Types::GetDeviceFleetReportResponse#device_stats #device_stats} => Types::DeviceStats
# * {Types::GetDeviceFleetReportResponse#agent_versions #agent_versions} => Array<Types::AgentVersion>
# * {Types::GetDeviceFleetReportResponse#model_stats #model_stats} => Array<Types::EdgeModelStat>
#
# @example Request syntax with placeholder values
#
# resp = client.get_device_fleet_report({
# device_fleet_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.device_fleet_arn #=> String
# resp.device_fleet_name #=> String
# resp.output_config.s3_output_location #=> String
# resp.output_config.kms_key_id #=> String
# resp.output_config.preset_deployment_type #=> String, one of "GreengrassV2Component"
# resp.output_config.preset_deployment_config #=> String
# resp.description #=> String
# resp.report_generated #=> Time
# resp.device_stats.connected_device_count #=> Integer
# resp.device_stats.registered_device_count #=> Integer
# resp.agent_versions #=> Array
# resp.agent_versions[0].version #=> String
# resp.agent_versions[0].agent_count #=> Integer
# resp.model_stats #=> Array
# resp.model_stats[0].model_name #=> String
# resp.model_stats[0].model_version #=> String
# resp.model_stats[0].offline_device_count #=> Integer
# resp.model_stats[0].connected_device_count #=> Integer
# resp.model_stats[0].active_device_count #=> Integer
# resp.model_stats[0].sampling_device_count #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/GetDeviceFleetReport AWS API Documentation
#
# @overload get_device_fleet_report(params = {})
# @param [Hash] params ({})
def get_device_fleet_report(params = {}, options = {})
req = build_request(:get_device_fleet_report, params)
req.send_request(options)
end
# The resource policy for the lineage group.
#
# @option params [required, String] :lineage_group_name
# The name or Amazon Resource Name (ARN) of the lineage group.
#
# @return [Types::GetLineageGroupPolicyResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::GetLineageGroupPolicyResponse#lineage_group_arn #lineage_group_arn} => String
# * {Types::GetLineageGroupPolicyResponse#resource_policy #resource_policy} => String
#
# @example Request syntax with placeholder values
#
# resp = client.get_lineage_group_policy({
# lineage_group_name: "LineageGroupNameOrArn", # required
# })
#
# @example Response structure
#
# resp.lineage_group_arn #=> String
# resp.resource_policy #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/GetLineageGroupPolicy AWS API Documentation
#
# @overload get_lineage_group_policy(params = {})
# @param [Hash] params ({})
def get_lineage_group_policy(params = {}, options = {})
req = build_request(:get_lineage_group_policy, params)
req.send_request(options)
end
# Gets a resource policy that manages access for a model group. For
# information about resource policies, see [Identity-based policies and
# resource-based policies][1] in the *Amazon Web Services Identity and
# Access Management User Guide.*.
#
#
#
# [1]: https://docs.aws.amazon.com/IAM/latest/UserGuide/access_policies_identity-vs-resource.html
#
# @option params [required, String] :model_package_group_name
# The name of the model group for which to get the resource policy.
#
# @return [Types::GetModelPackageGroupPolicyOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::GetModelPackageGroupPolicyOutput#resource_policy #resource_policy} => String
#
# @example Request syntax with placeholder values
#
# resp = client.get_model_package_group_policy({
# model_package_group_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.resource_policy #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/GetModelPackageGroupPolicy AWS API Documentation
#
# @overload get_model_package_group_policy(params = {})
# @param [Hash] params ({})
def get_model_package_group_policy(params = {}, options = {})
req = build_request(:get_model_package_group_policy, params)
req.send_request(options)
end
# Gets the status of Service Catalog in SageMaker. Service Catalog is
# used to create SageMaker projects.
#
# @return [Types::GetSagemakerServicecatalogPortfolioStatusOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::GetSagemakerServicecatalogPortfolioStatusOutput#status #status} => String
#
# @example Response structure
#
# resp.status #=> String, one of "Enabled", "Disabled"
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/GetSagemakerServicecatalogPortfolioStatus AWS API Documentation
#
# @overload get_sagemaker_servicecatalog_portfolio_status(params = {})
# @param [Hash] params ({})
def get_sagemaker_servicecatalog_portfolio_status(params = {}, options = {})
req = build_request(:get_sagemaker_servicecatalog_portfolio_status, params)
req.send_request(options)
end
# An auto-complete API for the search functionality in the SageMaker
# console. It returns suggestions of possible matches for the property
# name to use in `Search` queries. Provides suggestions for
# `HyperParameters`, `Tags`, and `Metrics`.
#
# @option params [required, String] :resource
# The name of the SageMaker resource to search for.
#
# @option params [Types::SuggestionQuery] :suggestion_query
# Limits the property names that are included in the response.
#
# @return [Types::GetSearchSuggestionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::GetSearchSuggestionsResponse#property_name_suggestions #property_name_suggestions} => Array<Types::PropertyNameSuggestion>
#
# @example Request syntax with placeholder values
#
# resp = client.get_search_suggestions({
# resource: "TrainingJob", # required, accepts TrainingJob, Experiment, ExperimentTrial, ExperimentTrialComponent, Endpoint, ModelPackage, ModelPackageGroup, Pipeline, PipelineExecution, FeatureGroup, Project, FeatureMetadata, HyperParameterTuningJob, ModelCard, Model
# suggestion_query: {
# property_name_query: {
# property_name_hint: "PropertyNameHint", # required
# },
# },
# })
#
# @example Response structure
#
# resp.property_name_suggestions #=> Array
# resp.property_name_suggestions[0].property_name #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/GetSearchSuggestions AWS API Documentation
#
# @overload get_search_suggestions(params = {})
# @param [Hash] params ({})
def get_search_suggestions(params = {}, options = {})
req = build_request(:get_search_suggestions, params)
req.send_request(options)
end
# Import hub content.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_content_name
# The name of the hub content to import.
#
# @option params [String] :hub_content_version
# The version of the hub content to import.
#
# @option params [required, String] :hub_content_type
# The type of hub content to import.
#
# @option params [required, String] :document_schema_version
# The version of the hub content schema to import.
#
# @option params [required, String] :hub_name
# The name of the hub to import content into.
#
# @option params [String] :hub_content_display_name
# The display name of the hub content to import.
#
# @option params [String] :hub_content_description
# A description of the hub content to import.
#
# @option params [String] :hub_content_markdown
# A string that provides a description of the hub content. This string
# can include links, tables, and standard markdown formating.
#
# @option params [required, String] :hub_content_document
# The hub content document that describes information about the hub
# content such as type, associated containers, scripts, and more.
#
# @option params [Array] :hub_content_search_keywords
# The searchable keywords of the hub content.
#
# @option params [Array] :tags
# Any tags associated with the hub content.
#
# @return [Types::ImportHubContentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ImportHubContentResponse#hub_arn #hub_arn} => String
# * {Types::ImportHubContentResponse#hub_content_arn #hub_content_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.import_hub_content({
# hub_content_name: "HubContentName", # required
# hub_content_version: "HubContentVersion",
# hub_content_type: "Model", # required, accepts Model, Notebook
# document_schema_version: "DocumentSchemaVersion", # required
# hub_name: "HubName", # required
# hub_content_display_name: "HubContentDisplayName",
# hub_content_description: "HubContentDescription",
# hub_content_markdown: "HubContentMarkdown",
# hub_content_document: "HubContentDocument", # required
# hub_content_search_keywords: ["HubSearchKeyword"],
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.hub_arn #=> String
# resp.hub_content_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ImportHubContent AWS API Documentation
#
# @overload import_hub_content(params = {})
# @param [Hash] params ({})
def import_hub_content(params = {}, options = {})
req = build_request(:import_hub_content, params)
req.send_request(options)
end
# Lists the actions in your account and their properties.
#
# @option params [String] :source_uri
# A filter that returns only actions with the specified source URI.
#
# @option params [String] :action_type
# A filter that returns only actions of the specified type.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only actions created on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only actions created on or before the specified
# time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [String] :next_token
# If the previous call to `ListActions` didn't return the full set of
# actions, the call returns a token for getting the next set of actions.
#
# @option params [Integer] :max_results
# The maximum number of actions to return in the response. The default
# value is 10.
#
# @return [Types::ListActionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListActionsResponse#action_summaries #action_summaries} => Array<Types::ActionSummary>
# * {Types::ListActionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_actions({
# source_uri: "SourceUri",
# action_type: "String256",
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.action_summaries #=> Array
# resp.action_summaries[0].action_arn #=> String
# resp.action_summaries[0].action_name #=> String
# resp.action_summaries[0].source.source_uri #=> String
# resp.action_summaries[0].source.source_type #=> String
# resp.action_summaries[0].source.source_id #=> String
# resp.action_summaries[0].action_type #=> String
# resp.action_summaries[0].status #=> String, one of "Unknown", "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.action_summaries[0].creation_time #=> Time
# resp.action_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListActions AWS API Documentation
#
# @overload list_actions(params = {})
# @param [Hash] params ({})
def list_actions(params = {}, options = {})
req = build_request(:list_actions, params)
req.send_request(options)
end
# Lists the machine learning algorithms that have been created.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only algorithms created after the specified time
# (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only algorithms created before the specified
# time (timestamp).
#
# @option params [Integer] :max_results
# The maximum number of algorithms to return in the response.
#
# @option params [String] :name_contains
# A string in the algorithm name. This filter returns only algorithms
# whose name contains the specified string.
#
# @option params [String] :next_token
# If the response to a previous `ListAlgorithms` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of
# algorithms, use the token in the next request.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for the results. The default is `Ascending`.
#
# @return [Types::ListAlgorithmsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListAlgorithmsOutput#algorithm_summary_list #algorithm_summary_list} => Array<Types::AlgorithmSummary>
# * {Types::ListAlgorithmsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_algorithms({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# max_results: 1,
# name_contains: "NameContains",
# next_token: "NextToken",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.algorithm_summary_list #=> Array
# resp.algorithm_summary_list[0].algorithm_name #=> String
# resp.algorithm_summary_list[0].algorithm_arn #=> String
# resp.algorithm_summary_list[0].algorithm_description #=> String
# resp.algorithm_summary_list[0].creation_time #=> Time
# resp.algorithm_summary_list[0].algorithm_status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Deleting"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListAlgorithms AWS API Documentation
#
# @overload list_algorithms(params = {})
# @param [Hash] params ({})
def list_algorithms(params = {}, options = {})
req = build_request(:list_algorithms, params)
req.send_request(options)
end
# Lists the aliases of a specified image or image version.
#
# @option params [required, String] :image_name
# The name of the image.
#
# @option params [String] :alias
# The alias of the image version.
#
# @option params [Integer] :version
# The version of the image. If image version is not specified, the
# aliases of all versions of the image are listed.
#
# @option params [Integer] :max_results
# The maximum number of aliases to return.
#
# @option params [String] :next_token
# If the previous call to `ListAliases` didn't return the full set of
# aliases, the call returns a token for retrieving the next set of
# aliases.
#
# @return [Types::ListAliasesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListAliasesResponse#sage_maker_image_version_aliases #sage_maker_image_version_aliases} => Array<String>
# * {Types::ListAliasesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_aliases({
# image_name: "ImageName", # required
# alias: "SageMakerImageVersionAlias",
# version: 1,
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.sage_maker_image_version_aliases #=> Array
# resp.sage_maker_image_version_aliases[0] #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListAliases AWS API Documentation
#
# @overload list_aliases(params = {})
# @param [Hash] params ({})
def list_aliases(params = {}, options = {})
req = build_request(:list_aliases, params)
req.send_request(options)
end
# Lists the AppImageConfigs in your account and their properties. The
# list can be filtered by creation time or modified time, and whether
# the AppImageConfig name contains a specified string.
#
# @option params [Integer] :max_results
# The maximum number of AppImageConfigs to return in the response. The
# default value is 10.
#
# @option params [String] :next_token
# If the previous call to `ListImages` didn't return the full set of
# AppImageConfigs, the call returns a token for getting the next set of
# AppImageConfigs.
#
# @option params [String] :name_contains
# A filter that returns only AppImageConfigs whose name contains the
# specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only AppImageConfigs created on or before the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only AppImageConfigs created on or after the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :modified_time_before
# A filter that returns only AppImageConfigs modified on or before the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :modified_time_after
# A filter that returns only AppImageConfigs modified on or after the
# specified time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @return [Types::ListAppImageConfigsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListAppImageConfigsResponse#next_token #next_token} => String
# * {Types::ListAppImageConfigsResponse#app_image_configs #app_image_configs} => Array<Types::AppImageConfigDetails>
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_app_image_configs({
# max_results: 1,
# next_token: "NextToken",
# name_contains: "AppImageConfigName",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# modified_time_before: Time.now,
# modified_time_after: Time.now,
# sort_by: "CreationTime", # accepts CreationTime, LastModifiedTime, Name
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.next_token #=> String
# resp.app_image_configs #=> Array
# resp.app_image_configs[0].app_image_config_arn #=> String
# resp.app_image_configs[0].app_image_config_name #=> String
# resp.app_image_configs[0].creation_time #=> Time
# resp.app_image_configs[0].last_modified_time #=> Time
# resp.app_image_configs[0].kernel_gateway_image_config.kernel_specs #=> Array
# resp.app_image_configs[0].kernel_gateway_image_config.kernel_specs[0].name #=> String
# resp.app_image_configs[0].kernel_gateway_image_config.kernel_specs[0].display_name #=> String
# resp.app_image_configs[0].kernel_gateway_image_config.file_system_config.mount_path #=> String
# resp.app_image_configs[0].kernel_gateway_image_config.file_system_config.default_uid #=> Integer
# resp.app_image_configs[0].kernel_gateway_image_config.file_system_config.default_gid #=> Integer
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListAppImageConfigs AWS API Documentation
#
# @overload list_app_image_configs(params = {})
# @param [Hash] params ({})
def list_app_image_configs(params = {}, options = {})
req = build_request(:list_app_image_configs, params)
req.send_request(options)
end
# Lists apps.
#
# @option params [String] :next_token
# If the previous response was truncated, you will receive this token.
# Use it in your next request to receive the next set of results.
#
# @option params [Integer] :max_results
# Returns a list up to a specified limit.
#
# @option params [String] :sort_order
# The sort order for the results. The default is Ascending.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# CreationTime.
#
# @option params [String] :domain_id_equals
# A parameter to search for the domain ID.
#
# @option params [String] :user_profile_name_equals
# A parameter to search by user profile name. If `SpaceNameEquals` is
# set, then this value cannot be set.
#
# @option params [String] :space_name_equals
# A parameter to search by space name. If `UserProfileNameEquals` is
# set, then this value cannot be set.
#
# @return [Types::ListAppsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListAppsResponse#apps #apps} => Array<Types::AppDetails>
# * {Types::ListAppsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_apps({
# next_token: "NextToken",
# max_results: 1,
# sort_order: "Ascending", # accepts Ascending, Descending
# sort_by: "CreationTime", # accepts CreationTime
# domain_id_equals: "DomainId",
# user_profile_name_equals: "UserProfileName",
# space_name_equals: "SpaceName",
# })
#
# @example Response structure
#
# resp.apps #=> Array
# resp.apps[0].domain_id #=> String
# resp.apps[0].user_profile_name #=> String
# resp.apps[0].app_type #=> String, one of "JupyterServer", "KernelGateway", "TensorBoard", "RStudioServerPro", "RSessionGateway"
# resp.apps[0].app_name #=> String
# resp.apps[0].status #=> String, one of "Deleted", "Deleting", "Failed", "InService", "Pending"
# resp.apps[0].creation_time #=> Time
# resp.apps[0].space_name #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListApps AWS API Documentation
#
# @overload list_apps(params = {})
# @param [Hash] params ({})
def list_apps(params = {}, options = {})
req = build_request(:list_apps, params)
req.send_request(options)
end
# Lists the artifacts in your account and their properties.
#
# @option params [String] :source_uri
# A filter that returns only artifacts with the specified source URI.
#
# @option params [String] :artifact_type
# A filter that returns only artifacts of the specified type.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only artifacts created on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only artifacts created on or before the
# specified time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [String] :next_token
# If the previous call to `ListArtifacts` didn't return the full set of
# artifacts, the call returns a token for getting the next set of
# artifacts.
#
# @option params [Integer] :max_results
# The maximum number of artifacts to return in the response. The default
# value is 10.
#
# @return [Types::ListArtifactsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListArtifactsResponse#artifact_summaries #artifact_summaries} => Array<Types::ArtifactSummary>
# * {Types::ListArtifactsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_artifacts({
# source_uri: "SourceUri",
# artifact_type: "String256",
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "CreationTime", # accepts CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.artifact_summaries #=> Array
# resp.artifact_summaries[0].artifact_arn #=> String
# resp.artifact_summaries[0].artifact_name #=> String
# resp.artifact_summaries[0].source.source_uri #=> String
# resp.artifact_summaries[0].source.source_types #=> Array
# resp.artifact_summaries[0].source.source_types[0].source_id_type #=> String, one of "MD5Hash", "S3ETag", "S3Version", "Custom"
# resp.artifact_summaries[0].source.source_types[0].value #=> String
# resp.artifact_summaries[0].artifact_type #=> String
# resp.artifact_summaries[0].creation_time #=> Time
# resp.artifact_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListArtifacts AWS API Documentation
#
# @overload list_artifacts(params = {})
# @param [Hash] params ({})
def list_artifacts(params = {}, options = {})
req = build_request(:list_artifacts, params)
req.send_request(options)
end
# Lists the associations in your account and their properties.
#
# @option params [String] :source_arn
# A filter that returns only associations with the specified source ARN.
#
# @option params [String] :destination_arn
# A filter that returns only associations with the specified destination
# Amazon Resource Name (ARN).
#
# @option params [String] :source_type
# A filter that returns only associations with the specified source
# type.
#
# @option params [String] :destination_type
# A filter that returns only associations with the specified destination
# type.
#
# @option params [String] :association_type
# A filter that returns only associations of the specified type.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only associations created on or after the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only associations created on or before the
# specified time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [String] :next_token
# If the previous call to `ListAssociations` didn't return the full set
# of associations, the call returns a token for getting the next set of
# associations.
#
# @option params [Integer] :max_results
# The maximum number of associations to return in the response. The
# default value is 10.
#
# @return [Types::ListAssociationsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListAssociationsResponse#association_summaries #association_summaries} => Array<Types::AssociationSummary>
# * {Types::ListAssociationsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_associations({
# source_arn: "AssociationEntityArn",
# destination_arn: "AssociationEntityArn",
# source_type: "String256",
# destination_type: "String256",
# association_type: "ContributedTo", # accepts ContributedTo, AssociatedWith, DerivedFrom, Produced
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "SourceArn", # accepts SourceArn, DestinationArn, SourceType, DestinationType, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.association_summaries #=> Array
# resp.association_summaries[0].source_arn #=> String
# resp.association_summaries[0].destination_arn #=> String
# resp.association_summaries[0].source_type #=> String
# resp.association_summaries[0].destination_type #=> String
# resp.association_summaries[0].association_type #=> String, one of "ContributedTo", "AssociatedWith", "DerivedFrom", "Produced"
# resp.association_summaries[0].source_name #=> String
# resp.association_summaries[0].destination_name #=> String
# resp.association_summaries[0].creation_time #=> Time
# resp.association_summaries[0].created_by.user_profile_arn #=> String
# resp.association_summaries[0].created_by.user_profile_name #=> String
# resp.association_summaries[0].created_by.domain_id #=> String
# resp.association_summaries[0].created_by.iam_identity.arn #=> String
# resp.association_summaries[0].created_by.iam_identity.principal_id #=> String
# resp.association_summaries[0].created_by.iam_identity.source_identity #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListAssociations AWS API Documentation
#
# @overload list_associations(params = {})
# @param [Hash] params ({})
def list_associations(params = {}, options = {})
req = build_request(:list_associations, params)
req.send_request(options)
end
# Request a list of jobs.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Request a list of jobs, using a filter for time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Request a list of jobs, using a filter for time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# Request a list of jobs, using a filter for time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# Request a list of jobs, using a filter for time.
#
# @option params [String] :name_contains
# Request a list of jobs, using a search filter for name.
#
# @option params [String] :status_equals
# Request a list of jobs, using a filter for status.
#
# @option params [String] :sort_order
# The sort order for the results. The default is `Descending`.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is `Name`.
#
# @option params [Integer] :max_results
# Request a list of jobs up to a specified limit.
#
# @option params [String] :next_token
# If the previous response was truncated, you receive this token. Use it
# in your next request to receive the next set of results.
#
# @return [Types::ListAutoMLJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListAutoMLJobsResponse#auto_ml_job_summaries #auto_ml_job_summaries} => Array<Types::AutoMLJobSummary>
# * {Types::ListAutoMLJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_auto_ml_jobs({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "AutoMLNameContains",
# status_equals: "Completed", # accepts Completed, InProgress, Failed, Stopped, Stopping
# sort_order: "Ascending", # accepts Ascending, Descending
# sort_by: "Name", # accepts Name, CreationTime, Status
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.auto_ml_job_summaries #=> Array
# resp.auto_ml_job_summaries[0].auto_ml_job_name #=> String
# resp.auto_ml_job_summaries[0].auto_ml_job_arn #=> String
# resp.auto_ml_job_summaries[0].auto_ml_job_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.auto_ml_job_summaries[0].auto_ml_job_secondary_status #=> String, one of "Starting", "AnalyzingData", "FeatureEngineering", "ModelTuning", "MaxCandidatesReached", "Failed", "Stopped", "MaxAutoMLJobRuntimeReached", "Stopping", "CandidateDefinitionsGenerated", "GeneratingExplainabilityReport", "Completed", "ExplainabilityError", "DeployingModel", "ModelDeploymentError", "GeneratingModelInsightsReport", "ModelInsightsError", "TrainingModels"
# resp.auto_ml_job_summaries[0].creation_time #=> Time
# resp.auto_ml_job_summaries[0].end_time #=> Time
# resp.auto_ml_job_summaries[0].last_modified_time #=> Time
# resp.auto_ml_job_summaries[0].failure_reason #=> String
# resp.auto_ml_job_summaries[0].partial_failure_reasons #=> Array
# resp.auto_ml_job_summaries[0].partial_failure_reasons[0].partial_failure_message #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListAutoMLJobs AWS API Documentation
#
# @overload list_auto_ml_jobs(params = {})
# @param [Hash] params ({})
def list_auto_ml_jobs(params = {}, options = {})
req = build_request(:list_auto_ml_jobs, params)
req.send_request(options)
end
# List the candidates created for the job.
#
# @option params [required, String] :auto_ml_job_name
# List the candidates created for the job by providing the job's name.
#
# @option params [String] :status_equals
# List the candidates for the job and filter by status.
#
# @option params [String] :candidate_name_equals
# List the candidates for the job and filter by candidate name.
#
# @option params [String] :sort_order
# The sort order for the results. The default is `Ascending`.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# `Descending`.
#
# @option params [Integer] :max_results
# List the job's candidates up to a specified limit.
#
# @option params [String] :next_token
# If the previous response was truncated, you receive this token. Use it
# in your next request to receive the next set of results.
#
# @return [Types::ListCandidatesForAutoMLJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListCandidatesForAutoMLJobResponse#candidates #candidates} => Array<Types::AutoMLCandidate>
# * {Types::ListCandidatesForAutoMLJobResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_candidates_for_auto_ml_job({
# auto_ml_job_name: "AutoMLJobName", # required
# status_equals: "Completed", # accepts Completed, InProgress, Failed, Stopped, Stopping
# candidate_name_equals: "CandidateName",
# sort_order: "Ascending", # accepts Ascending, Descending
# sort_by: "CreationTime", # accepts CreationTime, Status, FinalObjectiveMetricValue
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.candidates #=> Array
# resp.candidates[0].candidate_name #=> String
# resp.candidates[0].final_auto_ml_job_objective_metric.type #=> String, one of "Maximize", "Minimize"
# resp.candidates[0].final_auto_ml_job_objective_metric.metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.candidates[0].final_auto_ml_job_objective_metric.value #=> Float
# resp.candidates[0].final_auto_ml_job_objective_metric.standard_metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.candidates[0].objective_status #=> String, one of "Succeeded", "Pending", "Failed"
# resp.candidates[0].candidate_steps #=> Array
# resp.candidates[0].candidate_steps[0].candidate_step_type #=> String, one of "AWS::SageMaker::TrainingJob", "AWS::SageMaker::TransformJob", "AWS::SageMaker::ProcessingJob"
# resp.candidates[0].candidate_steps[0].candidate_step_arn #=> String
# resp.candidates[0].candidate_steps[0].candidate_step_name #=> String
# resp.candidates[0].candidate_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.candidates[0].inference_containers #=> Array
# resp.candidates[0].inference_containers[0].image #=> String
# resp.candidates[0].inference_containers[0].model_data_url #=> String
# resp.candidates[0].inference_containers[0].environment #=> Hash
# resp.candidates[0].inference_containers[0].environment["EnvironmentKey"] #=> String
# resp.candidates[0].creation_time #=> Time
# resp.candidates[0].end_time #=> Time
# resp.candidates[0].last_modified_time #=> Time
# resp.candidates[0].failure_reason #=> String
# resp.candidates[0].candidate_properties.candidate_artifact_locations.explainability #=> String
# resp.candidates[0].candidate_properties.candidate_artifact_locations.model_insights #=> String
# resp.candidates[0].candidate_properties.candidate_metrics #=> Array
# resp.candidates[0].candidate_properties.candidate_metrics[0].metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro"
# resp.candidates[0].candidate_properties.candidate_metrics[0].value #=> Float
# resp.candidates[0].candidate_properties.candidate_metrics[0].set #=> String, one of "Train", "Validation", "Test"
# resp.candidates[0].candidate_properties.candidate_metrics[0].standard_metric_name #=> String, one of "Accuracy", "MSE", "F1", "F1macro", "AUC", "RMSE", "MAE", "R2", "BalancedAccuracy", "Precision", "PrecisionMacro", "Recall", "RecallMacro", "LogLoss", "InferenceLatency"
# resp.candidates[0].inference_container_definitions #=> Hash
# resp.candidates[0].inference_container_definitions["AutoMLProcessingUnit"] #=> Array
# resp.candidates[0].inference_container_definitions["AutoMLProcessingUnit"][0].image #=> String
# resp.candidates[0].inference_container_definitions["AutoMLProcessingUnit"][0].model_data_url #=> String
# resp.candidates[0].inference_container_definitions["AutoMLProcessingUnit"][0].environment #=> Hash
# resp.candidates[0].inference_container_definitions["AutoMLProcessingUnit"][0].environment["EnvironmentKey"] #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListCandidatesForAutoMLJob AWS API Documentation
#
# @overload list_candidates_for_auto_ml_job(params = {})
# @param [Hash] params ({})
def list_candidates_for_auto_ml_job(params = {}, options = {})
req = build_request(:list_candidates_for_auto_ml_job, params)
req.send_request(options)
end
# Gets a list of the Git repositories in your account.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only Git repositories that were created after
# the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only Git repositories that were created before
# the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only Git repositories that were last modified
# after the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only Git repositories that were last modified
# before the specified time.
#
# @option params [Integer] :max_results
# The maximum number of Git repositories to return in the response.
#
# @option params [String] :name_contains
# A string in the Git repositories name. This filter returns only
# repositories whose name contains the specified string.
#
# @option params [String] :next_token
# If the result of a `ListCodeRepositoriesOutput` request was truncated,
# the response includes a `NextToken`. To get the next set of Git
# repositories, use the token in the next request.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `Name`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @return [Types::ListCodeRepositoriesOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListCodeRepositoriesOutput#code_repository_summary_list #code_repository_summary_list} => Array<Types::CodeRepositorySummary>
# * {Types::ListCodeRepositoriesOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_code_repositories({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# max_results: 1,
# name_contains: "CodeRepositoryNameContains",
# next_token: "NextToken",
# sort_by: "Name", # accepts Name, CreationTime, LastModifiedTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.code_repository_summary_list #=> Array
# resp.code_repository_summary_list[0].code_repository_name #=> String
# resp.code_repository_summary_list[0].code_repository_arn #=> String
# resp.code_repository_summary_list[0].creation_time #=> Time
# resp.code_repository_summary_list[0].last_modified_time #=> Time
# resp.code_repository_summary_list[0].git_config.repository_url #=> String
# resp.code_repository_summary_list[0].git_config.branch #=> String
# resp.code_repository_summary_list[0].git_config.secret_arn #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListCodeRepositories AWS API Documentation
#
# @overload list_code_repositories(params = {})
# @param [Hash] params ({})
def list_code_repositories(params = {}, options = {})
req = build_request(:list_code_repositories, params)
req.send_request(options)
end
# Lists model compilation jobs that satisfy various filters.
#
# To create a model compilation job, use [CreateCompilationJob][1]. To
# get information about a particular model compilation job you have
# created, use [DescribeCompilationJob][2].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateCompilationJob.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeCompilationJob.html
#
# @option params [String] :next_token
# If the result of the previous `ListCompilationJobs` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of model compilation jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of model compilation jobs to return in the
# response.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns the model compilation jobs that were created
# after a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns the model compilation jobs that were created
# before a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns the model compilation jobs that were modified
# after a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns the model compilation jobs that were modified
# before a specified time.
#
# @option params [String] :name_contains
# A filter that returns the model compilation jobs whose name contains a
# specified string.
#
# @option params [String] :status_equals
# A filter that retrieves model compilation jobs with a specific
# `CompilationJobStatus` status.
#
# @option params [String] :sort_by
# The field by which to sort results. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @return [Types::ListCompilationJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListCompilationJobsResponse#compilation_job_summaries #compilation_job_summaries} => Array<Types::CompilationJobSummary>
# * {Types::ListCompilationJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_compilation_jobs({
# next_token: "NextToken",
# max_results: 1,
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# status_equals: "INPROGRESS", # accepts INPROGRESS, COMPLETED, FAILED, STARTING, STOPPING, STOPPED
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.compilation_job_summaries #=> Array
# resp.compilation_job_summaries[0].compilation_job_name #=> String
# resp.compilation_job_summaries[0].compilation_job_arn #=> String
# resp.compilation_job_summaries[0].creation_time #=> Time
# resp.compilation_job_summaries[0].compilation_start_time #=> Time
# resp.compilation_job_summaries[0].compilation_end_time #=> Time
# resp.compilation_job_summaries[0].compilation_target_device #=> String, one of "lambda", "ml_m4", "ml_m5", "ml_c4", "ml_c5", "ml_p2", "ml_p3", "ml_g4dn", "ml_inf1", "ml_eia2", "jetson_tx1", "jetson_tx2", "jetson_nano", "jetson_xavier", "rasp3b", "imx8qm", "deeplens", "rk3399", "rk3288", "aisage", "sbe_c", "qcs605", "qcs603", "sitara_am57x", "amba_cv2", "amba_cv22", "amba_cv25", "x86_win32", "x86_win64", "coreml", "jacinto_tda4vm", "imx8mplus"
# resp.compilation_job_summaries[0].compilation_target_platform_os #=> String, one of "ANDROID", "LINUX"
# resp.compilation_job_summaries[0].compilation_target_platform_arch #=> String, one of "X86_64", "X86", "ARM64", "ARM_EABI", "ARM_EABIHF"
# resp.compilation_job_summaries[0].compilation_target_platform_accelerator #=> String, one of "INTEL_GRAPHICS", "MALI", "NVIDIA", "NNA"
# resp.compilation_job_summaries[0].last_modified_time #=> Time
# resp.compilation_job_summaries[0].compilation_job_status #=> String, one of "INPROGRESS", "COMPLETED", "FAILED", "STARTING", "STOPPING", "STOPPED"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListCompilationJobs AWS API Documentation
#
# @overload list_compilation_jobs(params = {})
# @param [Hash] params ({})
def list_compilation_jobs(params = {}, options = {})
req = build_request(:list_compilation_jobs, params)
req.send_request(options)
end
# Lists the contexts in your account and their properties.
#
# @option params [String] :source_uri
# A filter that returns only contexts with the specified source URI.
#
# @option params [String] :context_type
# A filter that returns only contexts of the specified type.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only contexts created on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only contexts created on or before the specified
# time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [String] :next_token
# If the previous call to `ListContexts` didn't return the full set of
# contexts, the call returns a token for getting the next set of
# contexts.
#
# @option params [Integer] :max_results
# The maximum number of contexts to return in the response. The default
# value is 10.
#
# @return [Types::ListContextsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListContextsResponse#context_summaries #context_summaries} => Array<Types::ContextSummary>
# * {Types::ListContextsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_contexts({
# source_uri: "SourceUri",
# context_type: "String256",
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.context_summaries #=> Array
# resp.context_summaries[0].context_arn #=> String
# resp.context_summaries[0].context_name #=> String
# resp.context_summaries[0].source.source_uri #=> String
# resp.context_summaries[0].source.source_type #=> String
# resp.context_summaries[0].source.source_id #=> String
# resp.context_summaries[0].context_type #=> String
# resp.context_summaries[0].creation_time #=> Time
# resp.context_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListContexts AWS API Documentation
#
# @overload list_contexts(params = {})
# @param [Hash] params ({})
def list_contexts(params = {}, options = {})
req = build_request(:list_contexts, params)
req.send_request(options)
end
# Lists the data quality job definitions in your account.
#
# @option params [String] :endpoint_name
# A filter that lists the data quality job definitions associated with
# the specified endpoint.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Descending`.
#
# @option params [String] :next_token
# If the result of the previous `ListDataQualityJobDefinitions` request
# was truncated, the response includes a `NextToken`. To retrieve the
# next set of transform jobs, use the token in the next request.>
#
# @option params [Integer] :max_results
# The maximum number of data quality monitoring job definitions to
# return in the response.
#
# @option params [String] :name_contains
# A string in the data quality monitoring job definition name. This
# filter returns only data quality monitoring job definitions whose name
# contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only data quality monitoring job definitions
# created before the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only data quality monitoring job definitions
# created after the specified time.
#
# @return [Types::ListDataQualityJobDefinitionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListDataQualityJobDefinitionsResponse#job_definition_summaries #job_definition_summaries} => Array<Types::MonitoringJobDefinitionSummary>
# * {Types::ListDataQualityJobDefinitionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_data_quality_job_definitions({
# endpoint_name: "EndpointName",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# name_contains: "NameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.job_definition_summaries #=> Array
# resp.job_definition_summaries[0].monitoring_job_definition_name #=> String
# resp.job_definition_summaries[0].monitoring_job_definition_arn #=> String
# resp.job_definition_summaries[0].creation_time #=> Time
# resp.job_definition_summaries[0].endpoint_name #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListDataQualityJobDefinitions AWS API Documentation
#
# @overload list_data_quality_job_definitions(params = {})
# @param [Hash] params ({})
def list_data_quality_job_definitions(params = {}, options = {})
req = build_request(:list_data_quality_job_definitions, params)
req.send_request(options)
end
# Returns a list of devices in the fleet.
#
# @option params [String] :next_token
# The response from the last list when returning a list large enough to
# need tokening.
#
# @option params [Integer] :max_results
# The maximum number of results to select.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Filter fleets where packaging job was created after specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Filter fleets where the edge packaging job was created before
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# Select fleets where the job was updated after X
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# Select fleets where the job was updated before X
#
# @option params [String] :name_contains
# Filter for fleets containing this name in their fleet device name.
#
# @option params [String] :sort_by
# The column to sort by.
#
# @option params [String] :sort_order
# What direction to sort in.
#
# @return [Types::ListDeviceFleetsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListDeviceFleetsResponse#device_fleet_summaries #device_fleet_summaries} => Array<Types::DeviceFleetSummary>
# * {Types::ListDeviceFleetsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_device_fleets({
# next_token: "NextToken",
# max_results: 1,
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# sort_by: "NAME", # accepts NAME, CREATION_TIME, LAST_MODIFIED_TIME
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.device_fleet_summaries #=> Array
# resp.device_fleet_summaries[0].device_fleet_arn #=> String
# resp.device_fleet_summaries[0].device_fleet_name #=> String
# resp.device_fleet_summaries[0].creation_time #=> Time
# resp.device_fleet_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListDeviceFleets AWS API Documentation
#
# @overload list_device_fleets(params = {})
# @param [Hash] params ({})
def list_device_fleets(params = {}, options = {})
req = build_request(:list_device_fleets, params)
req.send_request(options)
end
# A list of devices.
#
# @option params [String] :next_token
# The response from the last list when returning a list large enough to
# need tokening.
#
# @option params [Integer] :max_results
# Maximum number of results to select.
#
# @option params [Time,DateTime,Date,Integer,String] :latest_heartbeat_after
# Select fleets where the job was updated after X
#
# @option params [String] :model_name
# A filter that searches devices that contains this name in any of their
# models.
#
# @option params [String] :device_fleet_name
# Filter for fleets containing this name in their device fleet name.
#
# @return [Types::ListDevicesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListDevicesResponse#device_summaries #device_summaries} => Array<Types::DeviceSummary>
# * {Types::ListDevicesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_devices({
# next_token: "NextToken",
# max_results: 1,
# latest_heartbeat_after: Time.now,
# model_name: "EntityName",
# device_fleet_name: "EntityName",
# })
#
# @example Response structure
#
# resp.device_summaries #=> Array
# resp.device_summaries[0].device_name #=> String
# resp.device_summaries[0].device_arn #=> String
# resp.device_summaries[0].description #=> String
# resp.device_summaries[0].device_fleet_name #=> String
# resp.device_summaries[0].iot_thing_name #=> String
# resp.device_summaries[0].registration_time #=> Time
# resp.device_summaries[0].latest_heartbeat #=> Time
# resp.device_summaries[0].models #=> Array
# resp.device_summaries[0].models[0].model_name #=> String
# resp.device_summaries[0].models[0].model_version #=> String
# resp.device_summaries[0].agent_version #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListDevices AWS API Documentation
#
# @overload list_devices(params = {})
# @param [Hash] params ({})
def list_devices(params = {}, options = {})
req = build_request(:list_devices, params)
req.send_request(options)
end
# Lists the domains.
#
# @option params [String] :next_token
# If the previous response was truncated, you will receive this token.
# Use it in your next request to receive the next set of results.
#
# @option params [Integer] :max_results
# Returns a list up to a specified limit.
#
# @return [Types::ListDomainsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListDomainsResponse#domains #domains} => Array<Types::DomainDetails>
# * {Types::ListDomainsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_domains({
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.domains #=> Array
# resp.domains[0].domain_arn #=> String
# resp.domains[0].domain_id #=> String
# resp.domains[0].domain_name #=> String
# resp.domains[0].status #=> String, one of "Deleting", "Failed", "InService", "Pending", "Updating", "Update_Failed", "Delete_Failed"
# resp.domains[0].creation_time #=> Time
# resp.domains[0].last_modified_time #=> Time
# resp.domains[0].url #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListDomains AWS API Documentation
#
# @overload list_domains(params = {})
# @param [Hash] params ({})
def list_domains(params = {}, options = {})
req = build_request(:list_domains, params)
req.send_request(options)
end
# Lists all edge deployment plans.
#
# @option params [String] :next_token
# The response from the last list when returning a list large enough to
# need tokening.
#
# @option params [Integer] :max_results
# The maximum number of results to select (50 by default).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Selects edge deployment plans created after this time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Selects edge deployment plans created before this time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# Selects edge deployment plans that were last updated after this time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# Selects edge deployment plans that were last updated before this time.
#
# @option params [String] :name_contains
# Selects edge deployment plans with names containing this name.
#
# @option params [String] :device_fleet_name_contains
# Selects edge deployment plans with a device fleet name containing this
# name.
#
# @option params [String] :sort_by
# The column by which to sort the edge deployment plans. Can be one of
# `NAME`, `DEVICEFLEETNAME`, `CREATIONTIME`, `LASTMODIFIEDTIME`.
#
# @option params [String] :sort_order
# The direction of the sorting (ascending or descending).
#
# @return [Types::ListEdgeDeploymentPlansResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListEdgeDeploymentPlansResponse#edge_deployment_plan_summaries #edge_deployment_plan_summaries} => Array<Types::EdgeDeploymentPlanSummary>
# * {Types::ListEdgeDeploymentPlansResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_edge_deployment_plans({
# next_token: "NextToken",
# max_results: 1,
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# device_fleet_name_contains: "NameContains",
# sort_by: "NAME", # accepts NAME, DEVICE_FLEET_NAME, CREATION_TIME, LAST_MODIFIED_TIME
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.edge_deployment_plan_summaries #=> Array
# resp.edge_deployment_plan_summaries[0].edge_deployment_plan_arn #=> String
# resp.edge_deployment_plan_summaries[0].edge_deployment_plan_name #=> String
# resp.edge_deployment_plan_summaries[0].device_fleet_name #=> String
# resp.edge_deployment_plan_summaries[0].edge_deployment_success #=> Integer
# resp.edge_deployment_plan_summaries[0].edge_deployment_pending #=> Integer
# resp.edge_deployment_plan_summaries[0].edge_deployment_failed #=> Integer
# resp.edge_deployment_plan_summaries[0].creation_time #=> Time
# resp.edge_deployment_plan_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListEdgeDeploymentPlans AWS API Documentation
#
# @overload list_edge_deployment_plans(params = {})
# @param [Hash] params ({})
def list_edge_deployment_plans(params = {}, options = {})
req = build_request(:list_edge_deployment_plans, params)
req.send_request(options)
end
# Returns a list of edge packaging jobs.
#
# @option params [String] :next_token
# The response from the last list when returning a list large enough to
# need tokening.
#
# @option params [Integer] :max_results
# Maximum number of results to select.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Select jobs where the job was created after specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Select jobs where the job was created before specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# Select jobs where the job was updated after specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# Select jobs where the job was updated before specified time.
#
# @option params [String] :name_contains
# Filter for jobs containing this name in their packaging job name.
#
# @option params [String] :model_name_contains
# Filter for jobs where the model name contains this string.
#
# @option params [String] :status_equals
# The job status to filter for.
#
# @option params [String] :sort_by
# Use to specify what column to sort by.
#
# @option params [String] :sort_order
# What direction to sort by.
#
# @return [Types::ListEdgePackagingJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListEdgePackagingJobsResponse#edge_packaging_job_summaries #edge_packaging_job_summaries} => Array<Types::EdgePackagingJobSummary>
# * {Types::ListEdgePackagingJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_edge_packaging_jobs({
# next_token: "NextToken",
# max_results: 1,
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# model_name_contains: "NameContains",
# status_equals: "STARTING", # accepts STARTING, INPROGRESS, COMPLETED, FAILED, STOPPING, STOPPED
# sort_by: "NAME", # accepts NAME, MODEL_NAME, CREATION_TIME, LAST_MODIFIED_TIME, STATUS
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.edge_packaging_job_summaries #=> Array
# resp.edge_packaging_job_summaries[0].edge_packaging_job_arn #=> String
# resp.edge_packaging_job_summaries[0].edge_packaging_job_name #=> String
# resp.edge_packaging_job_summaries[0].edge_packaging_job_status #=> String, one of "STARTING", "INPROGRESS", "COMPLETED", "FAILED", "STOPPING", "STOPPED"
# resp.edge_packaging_job_summaries[0].compilation_job_name #=> String
# resp.edge_packaging_job_summaries[0].model_name #=> String
# resp.edge_packaging_job_summaries[0].model_version #=> String
# resp.edge_packaging_job_summaries[0].creation_time #=> Time
# resp.edge_packaging_job_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListEdgePackagingJobs AWS API Documentation
#
# @overload list_edge_packaging_jobs(params = {})
# @param [Hash] params ({})
def list_edge_packaging_jobs(params = {}, options = {})
req = build_request(:list_edge_packaging_jobs, params)
req.send_request(options)
end
# Lists endpoint configurations.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Descending`.
#
# @option params [String] :next_token
# If the result of the previous `ListEndpointConfig` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of endpoint configurations, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of training jobs to return in the response.
#
# @option params [String] :name_contains
# A string in the endpoint configuration name. This filter returns only
# endpoint configurations whose name contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only endpoint configurations created before the
# specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only endpoint configurations with a creation
# time greater than or equal to the specified time (timestamp).
#
# @return [Types::ListEndpointConfigsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListEndpointConfigsOutput#endpoint_configs #endpoint_configs} => Array<Types::EndpointConfigSummary>
# * {Types::ListEndpointConfigsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_endpoint_configs({
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "PaginationToken",
# max_results: 1,
# name_contains: "EndpointConfigNameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.endpoint_configs #=> Array
# resp.endpoint_configs[0].endpoint_config_name #=> String
# resp.endpoint_configs[0].endpoint_config_arn #=> String
# resp.endpoint_configs[0].creation_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListEndpointConfigs AWS API Documentation
#
# @overload list_endpoint_configs(params = {})
# @param [Hash] params ({})
def list_endpoint_configs(params = {}, options = {})
req = build_request(:list_endpoint_configs, params)
req.send_request(options)
end
# Lists endpoints.
#
# @option params [String] :sort_by
# Sorts the list of results. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Descending`.
#
# @option params [String] :next_token
# If the result of a `ListEndpoints` request was truncated, the response
# includes a `NextToken`. To retrieve the next set of endpoints, use the
# token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of endpoints to return in the response. This value
# defaults to 10.
#
# @option params [String] :name_contains
# A string in endpoint names. This filter returns only endpoints whose
# name contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only endpoints that were created before the
# specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only endpoints with a creation time greater than
# or equal to the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only endpoints that were modified before the
# specified timestamp.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only endpoints that were modified after the
# specified timestamp.
#
# @option params [String] :status_equals
# A filter that returns only endpoints with the specified status.
#
# @return [Types::ListEndpointsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListEndpointsOutput#endpoints #endpoints} => Array<Types::EndpointSummary>
# * {Types::ListEndpointsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_endpoints({
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "PaginationToken",
# max_results: 1,
# name_contains: "EndpointNameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# last_modified_time_before: Time.now,
# last_modified_time_after: Time.now,
# status_equals: "OutOfService", # accepts OutOfService, Creating, Updating, SystemUpdating, RollingBack, InService, Deleting, Failed
# })
#
# @example Response structure
#
# resp.endpoints #=> Array
# resp.endpoints[0].endpoint_name #=> String
# resp.endpoints[0].endpoint_arn #=> String
# resp.endpoints[0].creation_time #=> Time
# resp.endpoints[0].last_modified_time #=> Time
# resp.endpoints[0].endpoint_status #=> String, one of "OutOfService", "Creating", "Updating", "SystemUpdating", "RollingBack", "InService", "Deleting", "Failed"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListEndpoints AWS API Documentation
#
# @overload list_endpoints(params = {})
# @param [Hash] params ({})
def list_endpoints(params = {}, options = {})
req = build_request(:list_endpoints, params)
req.send_request(options)
end
# Lists all the experiments in your account. The list can be filtered to
# show only experiments that were created in a specific time range. The
# list can be sorted by experiment name or creation time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only experiments created after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only experiments created before the specified
# time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [String] :next_token
# If the previous call to `ListExperiments` didn't return the full set
# of experiments, the call returns a token for getting the next set of
# experiments.
#
# @option params [Integer] :max_results
# The maximum number of experiments to return in the response. The
# default value is 10.
#
# @return [Types::ListExperimentsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListExperimentsResponse#experiment_summaries #experiment_summaries} => Array<Types::ExperimentSummary>
# * {Types::ListExperimentsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_experiments({
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.experiment_summaries #=> Array
# resp.experiment_summaries[0].experiment_arn #=> String
# resp.experiment_summaries[0].experiment_name #=> String
# resp.experiment_summaries[0].display_name #=> String
# resp.experiment_summaries[0].experiment_source.source_arn #=> String
# resp.experiment_summaries[0].experiment_source.source_type #=> String
# resp.experiment_summaries[0].creation_time #=> Time
# resp.experiment_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListExperiments AWS API Documentation
#
# @overload list_experiments(params = {})
# @param [Hash] params ({})
def list_experiments(params = {}, options = {})
req = build_request(:list_experiments, params)
req.send_request(options)
end
# List `FeatureGroup`s based on given filter and order.
#
# @option params [String] :name_contains
# A string that partially matches one or more `FeatureGroup`s names.
# Filters `FeatureGroup`s by name.
#
# @option params [String] :feature_group_status_equals
# A `FeatureGroup` status. Filters by `FeatureGroup` status.
#
# @option params [String] :offline_store_status_equals
# An `OfflineStore` status. Filters by `OfflineStore` status.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Use this parameter to search for `FeatureGroups`s created after a
# specific date and time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Use this parameter to search for `FeatureGroups`s created before a
# specific date and time.
#
# @option params [String] :sort_order
# The order in which feature groups are listed.
#
# @option params [String] :sort_by
# The value on which the feature group list is sorted.
#
# @option params [Integer] :max_results
# The maximum number of results returned by `ListFeatureGroups`.
#
# @option params [String] :next_token
# A token to resume pagination of `ListFeatureGroups` results.
#
# @return [Types::ListFeatureGroupsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListFeatureGroupsResponse#feature_group_summaries #feature_group_summaries} => Array<Types::FeatureGroupSummary>
# * {Types::ListFeatureGroupsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_feature_groups({
# name_contains: "FeatureGroupNameContains",
# feature_group_status_equals: "Creating", # accepts Creating, Created, CreateFailed, Deleting, DeleteFailed
# offline_store_status_equals: "Active", # accepts Active, Blocked, Disabled
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# sort_order: "Ascending", # accepts Ascending, Descending
# sort_by: "Name", # accepts Name, FeatureGroupStatus, OfflineStoreStatus, CreationTime
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.feature_group_summaries #=> Array
# resp.feature_group_summaries[0].feature_group_name #=> String
# resp.feature_group_summaries[0].feature_group_arn #=> String
# resp.feature_group_summaries[0].creation_time #=> Time
# resp.feature_group_summaries[0].feature_group_status #=> String, one of "Creating", "Created", "CreateFailed", "Deleting", "DeleteFailed"
# resp.feature_group_summaries[0].offline_store_status.status #=> String, one of "Active", "Blocked", "Disabled"
# resp.feature_group_summaries[0].offline_store_status.blocked_reason #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListFeatureGroups AWS API Documentation
#
# @overload list_feature_groups(params = {})
# @param [Hash] params ({})
def list_feature_groups(params = {}, options = {})
req = build_request(:list_feature_groups, params)
req.send_request(options)
end
# Returns information about the flow definitions in your account.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only flow definitions with a creation time
# greater than or equal to the specified timestamp.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only flow definitions that were created before
# the specified timestamp.
#
# @option params [String] :sort_order
# An optional value that specifies whether you want the results sorted
# in `Ascending` or `Descending` order.
#
# @option params [String] :next_token
# A token to resume pagination.
#
# @option params [Integer] :max_results
# The total number of items to return. If the total number of available
# items is more than the value specified in `MaxResults`, then a
# `NextToken` will be provided in the output that you can use to resume
# pagination.
#
# @return [Types::ListFlowDefinitionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListFlowDefinitionsResponse#flow_definition_summaries #flow_definition_summaries} => Array<Types::FlowDefinitionSummary>
# * {Types::ListFlowDefinitionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_flow_definitions({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.flow_definition_summaries #=> Array
# resp.flow_definition_summaries[0].flow_definition_name #=> String
# resp.flow_definition_summaries[0].flow_definition_arn #=> String
# resp.flow_definition_summaries[0].flow_definition_status #=> String, one of "Initializing", "Active", "Failed", "Deleting"
# resp.flow_definition_summaries[0].creation_time #=> Time
# resp.flow_definition_summaries[0].failure_reason #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListFlowDefinitions AWS API Documentation
#
# @overload list_flow_definitions(params = {})
# @param [Hash] params ({})
def list_flow_definitions(params = {}, options = {})
req = build_request(:list_flow_definitions, params)
req.send_request(options)
end
# List hub content versions.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub to list the content versions of.
#
# @option params [required, String] :hub_content_type
# The type of hub content to list versions of.
#
# @option params [required, String] :hub_content_name
# The name of the hub content.
#
# @option params [String] :min_version
# The lower bound of the hub content versions to list.
#
# @option params [String] :max_schema_version
# The upper bound of the hub content schema version.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Only list hub content versions that were created before the time
# specified.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Only list hub content versions that were created after the time
# specified.
#
# @option params [String] :sort_by
# Sort hub content versions by either name or creation time.
#
# @option params [String] :sort_order
# Sort hub content versions by ascending or descending order.
#
# @option params [Integer] :max_results
# The maximum number of hub content versions to list.
#
# @option params [String] :next_token
# If the response to a previous `ListHubContentVersions` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of hub content versions, use the token in the next request.
#
# @return [Types::ListHubContentVersionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListHubContentVersionsResponse#hub_content_summaries #hub_content_summaries} => Array<Types::HubContentInfo>
# * {Types::ListHubContentVersionsResponse#next_token #next_token} => String
#
# @example Request syntax with placeholder values
#
# resp = client.list_hub_content_versions({
# hub_name: "HubName", # required
# hub_content_type: "Model", # required, accepts Model, Notebook
# hub_content_name: "HubContentName", # required
# min_version: "HubContentVersion",
# max_schema_version: "DocumentSchemaVersion",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# sort_by: "HubContentName", # accepts HubContentName, CreationTime, HubContentStatus
# sort_order: "Ascending", # accepts Ascending, Descending
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.hub_content_summaries #=> Array
# resp.hub_content_summaries[0].hub_content_name #=> String
# resp.hub_content_summaries[0].hub_content_arn #=> String
# resp.hub_content_summaries[0].hub_content_version #=> String
# resp.hub_content_summaries[0].hub_content_type #=> String, one of "Model", "Notebook"
# resp.hub_content_summaries[0].document_schema_version #=> String
# resp.hub_content_summaries[0].hub_content_display_name #=> String
# resp.hub_content_summaries[0].hub_content_description #=> String
# resp.hub_content_summaries[0].hub_content_search_keywords #=> Array
# resp.hub_content_summaries[0].hub_content_search_keywords[0] #=> String
# resp.hub_content_summaries[0].hub_content_status #=> String, one of "Available", "Importing", "Deleting", "ImportFailed", "DeleteFailed"
# resp.hub_content_summaries[0].creation_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListHubContentVersions AWS API Documentation
#
# @overload list_hub_content_versions(params = {})
# @param [Hash] params ({})
def list_hub_content_versions(params = {}, options = {})
req = build_request(:list_hub_content_versions, params)
req.send_request(options)
end
# List the contents of a hub.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub to list the contents of.
#
# @option params [required, String] :hub_content_type
# The type of hub content to list.
#
# @option params [String] :name_contains
# Only list hub content if the name contains the specified string.
#
# @option params [String] :max_schema_version
# The upper bound of the hub content schema verion.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Only list hub content that was created before the time specified.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Only list hub content that was created after the time specified.
#
# @option params [String] :sort_by
# Sort hub content versions by either name or creation time.
#
# @option params [String] :sort_order
# Sort hubs by ascending or descending order.
#
# @option params [Integer] :max_results
# The maximum amount of hub content to list.
#
# @option params [String] :next_token
# If the response to a previous `ListHubContents` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of hub
# content, use the token in the next request.
#
# @return [Types::ListHubContentsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListHubContentsResponse#hub_content_summaries #hub_content_summaries} => Array<Types::HubContentInfo>
# * {Types::ListHubContentsResponse#next_token #next_token} => String
#
# @example Request syntax with placeholder values
#
# resp = client.list_hub_contents({
# hub_name: "HubName", # required
# hub_content_type: "Model", # required, accepts Model, Notebook
# name_contains: "NameContains",
# max_schema_version: "DocumentSchemaVersion",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# sort_by: "HubContentName", # accepts HubContentName, CreationTime, HubContentStatus
# sort_order: "Ascending", # accepts Ascending, Descending
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.hub_content_summaries #=> Array
# resp.hub_content_summaries[0].hub_content_name #=> String
# resp.hub_content_summaries[0].hub_content_arn #=> String
# resp.hub_content_summaries[0].hub_content_version #=> String
# resp.hub_content_summaries[0].hub_content_type #=> String, one of "Model", "Notebook"
# resp.hub_content_summaries[0].document_schema_version #=> String
# resp.hub_content_summaries[0].hub_content_display_name #=> String
# resp.hub_content_summaries[0].hub_content_description #=> String
# resp.hub_content_summaries[0].hub_content_search_keywords #=> Array
# resp.hub_content_summaries[0].hub_content_search_keywords[0] #=> String
# resp.hub_content_summaries[0].hub_content_status #=> String, one of "Available", "Importing", "Deleting", "ImportFailed", "DeleteFailed"
# resp.hub_content_summaries[0].creation_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListHubContents AWS API Documentation
#
# @overload list_hub_contents(params = {})
# @param [Hash] params ({})
def list_hub_contents(params = {}, options = {})
req = build_request(:list_hub_contents, params)
req.send_request(options)
end
# List all existing hubs.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [String] :name_contains
# Only list hubs with names that contain the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Only list hubs that were created before the time specified.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Only list hubs that were created after the time specified.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# Only list hubs that were last modified before the time specified.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# Only list hubs that were last modified after the time specified.
#
# @option params [String] :sort_by
# Sort hubs by either name or creation time.
#
# @option params [String] :sort_order
# Sort hubs by ascending or descending order.
#
# @option params [Integer] :max_results
# The maximum number of hubs to list.
#
# @option params [String] :next_token
# If the response to a previous `ListHubs` request was truncated, the
# response includes a `NextToken`. To retrieve the next set of hubs, use
# the token in the next request.
#
# @return [Types::ListHubsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListHubsResponse#hub_summaries #hub_summaries} => Array<Types::HubInfo>
# * {Types::ListHubsResponse#next_token #next_token} => String
#
# @example Request syntax with placeholder values
#
# resp = client.list_hubs({
# name_contains: "NameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# last_modified_time_before: Time.now,
# last_modified_time_after: Time.now,
# sort_by: "HubName", # accepts HubName, CreationTime, HubStatus, AccountIdOwner
# sort_order: "Ascending", # accepts Ascending, Descending
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.hub_summaries #=> Array
# resp.hub_summaries[0].hub_name #=> String
# resp.hub_summaries[0].hub_arn #=> String
# resp.hub_summaries[0].hub_display_name #=> String
# resp.hub_summaries[0].hub_description #=> String
# resp.hub_summaries[0].hub_search_keywords #=> Array
# resp.hub_summaries[0].hub_search_keywords[0] #=> String
# resp.hub_summaries[0].hub_status #=> String, one of "InService", "Creating", "Updating", "Deleting", "CreateFailed", "UpdateFailed", "DeleteFailed"
# resp.hub_summaries[0].creation_time #=> Time
# resp.hub_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListHubs AWS API Documentation
#
# @overload list_hubs(params = {})
# @param [Hash] params ({})
def list_hubs(params = {}, options = {})
req = build_request(:list_hubs, params)
req.send_request(options)
end
# Returns information about the human task user interfaces in your
# account.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only human task user interfaces with a creation
# time greater than or equal to the specified timestamp.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only human task user interfaces that were
# created before the specified timestamp.
#
# @option params [String] :sort_order
# An optional value that specifies whether you want the results sorted
# in `Ascending` or `Descending` order.
#
# @option params [String] :next_token
# A token to resume pagination.
#
# @option params [Integer] :max_results
# The total number of items to return. If the total number of available
# items is more than the value specified in `MaxResults`, then a
# `NextToken` will be provided in the output that you can use to resume
# pagination.
#
# @return [Types::ListHumanTaskUisResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListHumanTaskUisResponse#human_task_ui_summaries #human_task_ui_summaries} => Array<Types::HumanTaskUiSummary>
# * {Types::ListHumanTaskUisResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_human_task_uis({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.human_task_ui_summaries #=> Array
# resp.human_task_ui_summaries[0].human_task_ui_name #=> String
# resp.human_task_ui_summaries[0].human_task_ui_arn #=> String
# resp.human_task_ui_summaries[0].creation_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListHumanTaskUis AWS API Documentation
#
# @overload list_human_task_uis(params = {})
# @param [Hash] params ({})
def list_human_task_uis(params = {}, options = {})
req = build_request(:list_human_task_uis, params)
req.send_request(options)
end
# Gets a list of [HyperParameterTuningJobSummary][1] objects that
# describe the hyperparameter tuning jobs launched in your account.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_HyperParameterTuningJobSummary.html
#
# @option params [String] :next_token
# If the result of the previous `ListHyperParameterTuningJobs` request
# was truncated, the response includes a `NextToken`. To retrieve the
# next set of tuning jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of tuning jobs to return. The default value is 10.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `Name`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @option params [String] :name_contains
# A string in the tuning job name. This filter returns only tuning jobs
# whose name contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only tuning jobs that were created after the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only tuning jobs that were created before the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only tuning jobs that were modified after the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only tuning jobs that were modified before the
# specified time.
#
# @option params [String] :status_equals
# A filter that returns only tuning jobs with the specified status.
#
# @return [Types::ListHyperParameterTuningJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListHyperParameterTuningJobsResponse#hyper_parameter_tuning_job_summaries #hyper_parameter_tuning_job_summaries} => Array<Types::HyperParameterTuningJobSummary>
# * {Types::ListHyperParameterTuningJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_hyper_parameter_tuning_jobs({
# next_token: "NextToken",
# max_results: 1,
# sort_by: "Name", # accepts Name, Status, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# name_contains: "NameContains",
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# status_equals: "Completed", # accepts Completed, InProgress, Failed, Stopped, Stopping
# })
#
# @example Response structure
#
# resp.hyper_parameter_tuning_job_summaries #=> Array
# resp.hyper_parameter_tuning_job_summaries[0].hyper_parameter_tuning_job_name #=> String
# resp.hyper_parameter_tuning_job_summaries[0].hyper_parameter_tuning_job_arn #=> String
# resp.hyper_parameter_tuning_job_summaries[0].hyper_parameter_tuning_job_status #=> String, one of "Completed", "InProgress", "Failed", "Stopped", "Stopping"
# resp.hyper_parameter_tuning_job_summaries[0].strategy #=> String, one of "Bayesian", "Random", "Hyperband", "Grid"
# resp.hyper_parameter_tuning_job_summaries[0].creation_time #=> Time
# resp.hyper_parameter_tuning_job_summaries[0].hyper_parameter_tuning_end_time #=> Time
# resp.hyper_parameter_tuning_job_summaries[0].last_modified_time #=> Time
# resp.hyper_parameter_tuning_job_summaries[0].training_job_status_counters.completed #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].training_job_status_counters.in_progress #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].training_job_status_counters.retryable_error #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].training_job_status_counters.non_retryable_error #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].training_job_status_counters.stopped #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].objective_status_counters.succeeded #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].objective_status_counters.pending #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].objective_status_counters.failed #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].resource_limits.max_number_of_training_jobs #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].resource_limits.max_parallel_training_jobs #=> Integer
# resp.hyper_parameter_tuning_job_summaries[0].resource_limits.max_runtime_in_seconds #=> Integer
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListHyperParameterTuningJobs AWS API Documentation
#
# @overload list_hyper_parameter_tuning_jobs(params = {})
# @param [Hash] params ({})
def list_hyper_parameter_tuning_jobs(params = {}, options = {})
req = build_request(:list_hyper_parameter_tuning_jobs, params)
req.send_request(options)
end
# Lists the versions of a specified image and their properties. The list
# can be filtered by creation time or modified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only versions created on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only versions created on or before the specified
# time.
#
# @option params [required, String] :image_name
# The name of the image to list the versions of.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only versions modified on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only versions modified on or before the
# specified time.
#
# @option params [Integer] :max_results
# The maximum number of versions to return in the response. The default
# value is 10.
#
# @option params [String] :next_token
# If the previous call to `ListImageVersions` didn't return the full
# set of versions, the call returns a token for getting the next set of
# versions.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CREATION_TIME`.
#
# @option params [String] :sort_order
# The sort order. The default value is `DESCENDING`.
#
# @return [Types::ListImageVersionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListImageVersionsResponse#image_versions #image_versions} => Array<Types::ImageVersion>
# * {Types::ListImageVersionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_image_versions({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# image_name: "ImageName", # required
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# max_results: 1,
# next_token: "NextToken",
# sort_by: "CREATION_TIME", # accepts CREATION_TIME, LAST_MODIFIED_TIME, VERSION
# sort_order: "ASCENDING", # accepts ASCENDING, DESCENDING
# })
#
# @example Response structure
#
# resp.image_versions #=> Array
# resp.image_versions[0].creation_time #=> Time
# resp.image_versions[0].failure_reason #=> String
# resp.image_versions[0].image_arn #=> String
# resp.image_versions[0].image_version_arn #=> String
# resp.image_versions[0].image_version_status #=> String, one of "CREATING", "CREATED", "CREATE_FAILED", "DELETING", "DELETE_FAILED"
# resp.image_versions[0].last_modified_time #=> Time
# resp.image_versions[0].version #=> Integer
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListImageVersions AWS API Documentation
#
# @overload list_image_versions(params = {})
# @param [Hash] params ({})
def list_image_versions(params = {}, options = {})
req = build_request(:list_image_versions, params)
req.send_request(options)
end
# Lists the images in your account and their properties. The list can be
# filtered by creation time or modified time, and whether the image name
# contains a specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only images created on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only images created on or before the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only images modified on or after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only images modified on or before the specified
# time.
#
# @option params [Integer] :max_results
# The maximum number of images to return in the response. The default
# value is 10.
#
# @option params [String] :name_contains
# A filter that returns only images whose name contains the specified
# string.
#
# @option params [String] :next_token
# If the previous call to `ListImages` didn't return the full set of
# images, the call returns a token for getting the next set of images.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CREATION_TIME`.
#
# @option params [String] :sort_order
# The sort order. The default value is `DESCENDING`.
#
# @return [Types::ListImagesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListImagesResponse#images #images} => Array<Types::Image>
# * {Types::ListImagesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_images({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# max_results: 1,
# name_contains: "ImageNameContains",
# next_token: "NextToken",
# sort_by: "CREATION_TIME", # accepts CREATION_TIME, LAST_MODIFIED_TIME, IMAGE_NAME
# sort_order: "ASCENDING", # accepts ASCENDING, DESCENDING
# })
#
# @example Response structure
#
# resp.images #=> Array
# resp.images[0].creation_time #=> Time
# resp.images[0].description #=> String
# resp.images[0].display_name #=> String
# resp.images[0].failure_reason #=> String
# resp.images[0].image_arn #=> String
# resp.images[0].image_name #=> String
# resp.images[0].image_status #=> String, one of "CREATING", "CREATED", "CREATE_FAILED", "UPDATING", "UPDATE_FAILED", "DELETING", "DELETE_FAILED"
# resp.images[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListImages AWS API Documentation
#
# @overload list_images(params = {})
# @param [Hash] params ({})
def list_images(params = {}, options = {})
req = build_request(:list_images, params)
req.send_request(options)
end
# Returns the list of all inference experiments.
#
# @option params [String] :name_contains
# Selects inference experiments whose names contain this name.
#
# @option params [String] :type
# Selects inference experiments of this type. For the possible types of
# inference experiments, see [CreateInferenceExperiment][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateInferenceExperiment.html
#
# @option params [String] :status_equals
# Selects inference experiments which are in this status. For the
# possible statuses, see [DescribeInferenceExperiment][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeInferenceExperiment.html
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Selects inference experiments which were created after this timestamp.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Selects inference experiments which were created before this
# timestamp.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# Selects inference experiments which were last modified after this
# timestamp.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# Selects inference experiments which were last modified before this
# timestamp.
#
# @option params [String] :sort_by
# The column by which to sort the listed inference experiments.
#
# @option params [String] :sort_order
# The direction of sorting (ascending or descending).
#
# @option params [String] :next_token
# The response from the last list when returning a list large enough to
# need tokening.
#
# @option params [Integer] :max_results
# The maximum number of results to select.
#
# @return [Types::ListInferenceExperimentsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListInferenceExperimentsResponse#inference_experiments #inference_experiments} => Array<Types::InferenceExperimentSummary>
# * {Types::ListInferenceExperimentsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_inference_experiments({
# name_contains: "NameContains",
# type: "ShadowMode", # accepts ShadowMode
# status_equals: "Creating", # accepts Creating, Created, Updating, Running, Starting, Stopping, Completed, Cancelled
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.inference_experiments #=> Array
# resp.inference_experiments[0].name #=> String
# resp.inference_experiments[0].type #=> String, one of "ShadowMode"
# resp.inference_experiments[0].schedule.start_time #=> Time
# resp.inference_experiments[0].schedule.end_time #=> Time
# resp.inference_experiments[0].status #=> String, one of "Creating", "Created", "Updating", "Running", "Starting", "Stopping", "Completed", "Cancelled"
# resp.inference_experiments[0].status_reason #=> String
# resp.inference_experiments[0].description #=> String
# resp.inference_experiments[0].creation_time #=> Time
# resp.inference_experiments[0].completion_time #=> Time
# resp.inference_experiments[0].last_modified_time #=> Time
# resp.inference_experiments[0].role_arn #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListInferenceExperiments AWS API Documentation
#
# @overload list_inference_experiments(params = {})
# @param [Hash] params ({})
def list_inference_experiments(params = {}, options = {})
req = build_request(:list_inference_experiments, params)
req.send_request(options)
end
# Returns a list of the subtasks for an Inference Recommender job.
#
# The supported subtasks are benchmarks, which evaluate the performance
# of your model on different instance types.
#
# @option params [required, String] :job_name
# The name for the Inference Recommender job.
#
# @option params [String] :status
# A filter to return benchmarks of a specified status. If this field is
# left empty, then all benchmarks are returned.
#
# @option params [String] :step_type
# A filter to return details about the specified type of subtask.
#
# `BENCHMARK`: Evaluate the performance of your model on different
# instance types.
#
# @option params [Integer] :max_results
# The maximum number of results to return.
#
# @option params [String] :next_token
# A token that you can specify to return more results from the list.
# Specify this field if you have a token that was returned from a
# previous request.
#
# @return [Types::ListInferenceRecommendationsJobStepsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListInferenceRecommendationsJobStepsResponse#steps #steps} => Array<Types::InferenceRecommendationsJobStep>
# * {Types::ListInferenceRecommendationsJobStepsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_inference_recommendations_job_steps({
# job_name: "RecommendationJobName", # required
# status: "PENDING", # accepts PENDING, IN_PROGRESS, COMPLETED, FAILED, STOPPING, STOPPED
# step_type: "BENCHMARK", # accepts BENCHMARK
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.steps #=> Array
# resp.steps[0].step_type #=> String, one of "BENCHMARK"
# resp.steps[0].job_name #=> String
# resp.steps[0].status #=> String, one of "PENDING", "IN_PROGRESS", "COMPLETED", "FAILED", "STOPPING", "STOPPED"
# resp.steps[0].inference_benchmark.metrics.cost_per_hour #=> Float
# resp.steps[0].inference_benchmark.metrics.cost_per_inference #=> Float
# resp.steps[0].inference_benchmark.metrics.max_invocations #=> Integer
# resp.steps[0].inference_benchmark.metrics.model_latency #=> Integer
# resp.steps[0].inference_benchmark.metrics.cpu_utilization #=> Float
# resp.steps[0].inference_benchmark.metrics.memory_utilization #=> Float
# resp.steps[0].inference_benchmark.endpoint_configuration.endpoint_name #=> String
# resp.steps[0].inference_benchmark.endpoint_configuration.variant_name #=> String
# resp.steps[0].inference_benchmark.endpoint_configuration.instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.large", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.12xlarge", "ml.m5d.24xlarge", "ml.c4.large", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.c5.large", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.large", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.12xlarge", "ml.r5.24xlarge", "ml.r5d.large", "ml.r5d.xlarge", "ml.r5d.2xlarge", "ml.r5d.4xlarge", "ml.r5d.12xlarge", "ml.r5d.24xlarge", "ml.inf1.xlarge", "ml.inf1.2xlarge", "ml.inf1.6xlarge", "ml.inf1.24xlarge", "ml.c6i.large", "ml.c6i.xlarge", "ml.c6i.2xlarge", "ml.c6i.4xlarge", "ml.c6i.8xlarge", "ml.c6i.12xlarge", "ml.c6i.16xlarge", "ml.c6i.24xlarge", "ml.c6i.32xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.12xlarge", "ml.g5.16xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge", "ml.p4d.24xlarge", "ml.c7g.large", "ml.c7g.xlarge", "ml.c7g.2xlarge", "ml.c7g.4xlarge", "ml.c7g.8xlarge", "ml.c7g.12xlarge", "ml.c7g.16xlarge", "ml.m6g.large", "ml.m6g.xlarge", "ml.m6g.2xlarge", "ml.m6g.4xlarge", "ml.m6g.8xlarge", "ml.m6g.12xlarge", "ml.m6g.16xlarge", "ml.m6gd.large", "ml.m6gd.xlarge", "ml.m6gd.2xlarge", "ml.m6gd.4xlarge", "ml.m6gd.8xlarge", "ml.m6gd.12xlarge", "ml.m6gd.16xlarge", "ml.c6g.large", "ml.c6g.xlarge", "ml.c6g.2xlarge", "ml.c6g.4xlarge", "ml.c6g.8xlarge", "ml.c6g.12xlarge", "ml.c6g.16xlarge", "ml.c6gd.large", "ml.c6gd.xlarge", "ml.c6gd.2xlarge", "ml.c6gd.4xlarge", "ml.c6gd.8xlarge", "ml.c6gd.12xlarge", "ml.c6gd.16xlarge", "ml.c6gn.large", "ml.c6gn.xlarge", "ml.c6gn.2xlarge", "ml.c6gn.4xlarge", "ml.c6gn.8xlarge", "ml.c6gn.12xlarge", "ml.c6gn.16xlarge", "ml.r6g.large", "ml.r6g.xlarge", "ml.r6g.2xlarge", "ml.r6g.4xlarge", "ml.r6g.8xlarge", "ml.r6g.12xlarge", "ml.r6g.16xlarge", "ml.r6gd.large", "ml.r6gd.xlarge", "ml.r6gd.2xlarge", "ml.r6gd.4xlarge", "ml.r6gd.8xlarge", "ml.r6gd.12xlarge", "ml.r6gd.16xlarge", "ml.p4de.24xlarge", "ml.trn1.2xlarge", "ml.trn1.32xlarge", "ml.inf2.xlarge", "ml.inf2.8xlarge", "ml.inf2.24xlarge", "ml.inf2.48xlarge"
# resp.steps[0].inference_benchmark.endpoint_configuration.initial_instance_count #=> Integer
# resp.steps[0].inference_benchmark.model_configuration.inference_specification_name #=> String
# resp.steps[0].inference_benchmark.model_configuration.environment_parameters #=> Array
# resp.steps[0].inference_benchmark.model_configuration.environment_parameters[0].key #=> String
# resp.steps[0].inference_benchmark.model_configuration.environment_parameters[0].value_type #=> String
# resp.steps[0].inference_benchmark.model_configuration.environment_parameters[0].value #=> String
# resp.steps[0].inference_benchmark.model_configuration.compilation_job_name #=> String
# resp.steps[0].inference_benchmark.failure_reason #=> String
# resp.steps[0].inference_benchmark.endpoint_metrics.max_invocations #=> Integer
# resp.steps[0].inference_benchmark.endpoint_metrics.model_latency #=> Integer
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListInferenceRecommendationsJobSteps AWS API Documentation
#
# @overload list_inference_recommendations_job_steps(params = {})
# @param [Hash] params ({})
def list_inference_recommendations_job_steps(params = {}, options = {})
req = build_request(:list_inference_recommendations_job_steps, params)
req.send_request(options)
end
# Lists recommendation jobs that satisfy various filters.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only jobs created after the specified time
# (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only jobs created before the specified time
# (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only jobs that were last modified after the
# specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only jobs that were last modified before the
# specified time (timestamp).
#
# @option params [String] :name_contains
# A string in the job name. This filter returns only recommendations
# whose name contains the specified string.
#
# @option params [String] :status_equals
# A filter that retrieves only inference recommendations jobs with a
# specific status.
#
# @option params [String] :sort_by
# The parameter by which to sort the results.
#
# @option params [String] :sort_order
# The sort order for the results.
#
# @option params [String] :next_token
# If the response to a previous
# `ListInferenceRecommendationsJobsRequest` request was truncated, the
# response includes a `NextToken`. To retrieve the next set of
# recommendations, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of recommendations to return in the response.
#
# @return [Types::ListInferenceRecommendationsJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListInferenceRecommendationsJobsResponse#inference_recommendations_jobs #inference_recommendations_jobs} => Array<Types::InferenceRecommendationsJob>
# * {Types::ListInferenceRecommendationsJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_inference_recommendations_jobs({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# status_equals: "PENDING", # accepts PENDING, IN_PROGRESS, COMPLETED, FAILED, STOPPING, STOPPED
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.inference_recommendations_jobs #=> Array
# resp.inference_recommendations_jobs[0].job_name #=> String
# resp.inference_recommendations_jobs[0].job_description #=> String
# resp.inference_recommendations_jobs[0].job_type #=> String, one of "Default", "Advanced"
# resp.inference_recommendations_jobs[0].job_arn #=> String
# resp.inference_recommendations_jobs[0].status #=> String, one of "PENDING", "IN_PROGRESS", "COMPLETED", "FAILED", "STOPPING", "STOPPED"
# resp.inference_recommendations_jobs[0].creation_time #=> Time
# resp.inference_recommendations_jobs[0].completion_time #=> Time
# resp.inference_recommendations_jobs[0].role_arn #=> String
# resp.inference_recommendations_jobs[0].last_modified_time #=> Time
# resp.inference_recommendations_jobs[0].failure_reason #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListInferenceRecommendationsJobs AWS API Documentation
#
# @overload list_inference_recommendations_jobs(params = {})
# @param [Hash] params ({})
def list_inference_recommendations_jobs(params = {}, options = {})
req = build_request(:list_inference_recommendations_jobs, params)
req.send_request(options)
end
# Gets a list of labeling jobs.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only labeling jobs created after the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only labeling jobs created before the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only labeling jobs modified after the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only labeling jobs modified before the specified
# time (timestamp).
#
# @option params [Integer] :max_results
# The maximum number of labeling jobs to return in each page of the
# response.
#
# @option params [String] :next_token
# If the result of the previous `ListLabelingJobs` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of labeling jobs, use the token in the next request.
#
# @option params [String] :name_contains
# A string in the labeling job name. This filter returns only labeling
# jobs whose name contains the specified string.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @option params [String] :status_equals
# A filter that retrieves only labeling jobs with a specific status.
#
# @return [Types::ListLabelingJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListLabelingJobsResponse#labeling_job_summary_list #labeling_job_summary_list} => Array<Types::LabelingJobSummary>
# * {Types::ListLabelingJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_labeling_jobs({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# max_results: 1,
# next_token: "NextToken",
# name_contains: "NameContains",
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# status_equals: "Initializing", # accepts Initializing, InProgress, Completed, Failed, Stopping, Stopped
# })
#
# @example Response structure
#
# resp.labeling_job_summary_list #=> Array
# resp.labeling_job_summary_list[0].labeling_job_name #=> String
# resp.labeling_job_summary_list[0].labeling_job_arn #=> String
# resp.labeling_job_summary_list[0].creation_time #=> Time
# resp.labeling_job_summary_list[0].last_modified_time #=> Time
# resp.labeling_job_summary_list[0].labeling_job_status #=> String, one of "Initializing", "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.labeling_job_summary_list[0].label_counters.total_labeled #=> Integer
# resp.labeling_job_summary_list[0].label_counters.human_labeled #=> Integer
# resp.labeling_job_summary_list[0].label_counters.machine_labeled #=> Integer
# resp.labeling_job_summary_list[0].label_counters.failed_non_retryable_error #=> Integer
# resp.labeling_job_summary_list[0].label_counters.unlabeled #=> Integer
# resp.labeling_job_summary_list[0].workteam_arn #=> String
# resp.labeling_job_summary_list[0].pre_human_task_lambda_arn #=> String
# resp.labeling_job_summary_list[0].annotation_consolidation_lambda_arn #=> String
# resp.labeling_job_summary_list[0].failure_reason #=> String
# resp.labeling_job_summary_list[0].labeling_job_output.output_dataset_s3_uri #=> String
# resp.labeling_job_summary_list[0].labeling_job_output.final_active_learning_model_arn #=> String
# resp.labeling_job_summary_list[0].input_config.data_source.s3_data_source.manifest_s3_uri #=> String
# resp.labeling_job_summary_list[0].input_config.data_source.sns_data_source.sns_topic_arn #=> String
# resp.labeling_job_summary_list[0].input_config.data_attributes.content_classifiers #=> Array
# resp.labeling_job_summary_list[0].input_config.data_attributes.content_classifiers[0] #=> String, one of "FreeOfPersonallyIdentifiableInformation", "FreeOfAdultContent"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListLabelingJobs AWS API Documentation
#
# @overload list_labeling_jobs(params = {})
# @param [Hash] params ({})
def list_labeling_jobs(params = {}, options = {})
req = build_request(:list_labeling_jobs, params)
req.send_request(options)
end
# Gets a list of labeling jobs assigned to a specified work team.
#
# @option params [required, String] :workteam_arn
# The Amazon Resource Name (ARN) of the work team for which you want to
# see labeling jobs for.
#
# @option params [Integer] :max_results
# The maximum number of labeling jobs to return in each page of the
# response.
#
# @option params [String] :next_token
# If the result of the previous `ListLabelingJobsForWorkteam` request
# was truncated, the response includes a `NextToken`. To retrieve the
# next set of labeling jobs, use the token in the next request.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only labeling jobs created after the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only labeling jobs created before the specified
# time (timestamp).
#
# @option params [String] :job_reference_code_contains
# A filter the limits jobs to only the ones whose job reference code
# contains the specified string.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @return [Types::ListLabelingJobsForWorkteamResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListLabelingJobsForWorkteamResponse#labeling_job_summary_list #labeling_job_summary_list} => Array<Types::LabelingJobForWorkteamSummary>
# * {Types::ListLabelingJobsForWorkteamResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_labeling_jobs_for_workteam({
# workteam_arn: "WorkteamArn", # required
# max_results: 1,
# next_token: "NextToken",
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# job_reference_code_contains: "JobReferenceCodeContains",
# sort_by: "CreationTime", # accepts CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.labeling_job_summary_list #=> Array
# resp.labeling_job_summary_list[0].labeling_job_name #=> String
# resp.labeling_job_summary_list[0].job_reference_code #=> String
# resp.labeling_job_summary_list[0].work_requester_account_id #=> String
# resp.labeling_job_summary_list[0].creation_time #=> Time
# resp.labeling_job_summary_list[0].label_counters.human_labeled #=> Integer
# resp.labeling_job_summary_list[0].label_counters.pending_human #=> Integer
# resp.labeling_job_summary_list[0].label_counters.total #=> Integer
# resp.labeling_job_summary_list[0].number_of_human_workers_per_data_object #=> Integer
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListLabelingJobsForWorkteam AWS API Documentation
#
# @overload list_labeling_jobs_for_workteam(params = {})
# @param [Hash] params ({})
def list_labeling_jobs_for_workteam(params = {}, options = {})
req = build_request(:list_labeling_jobs_for_workteam, params)
req.send_request(options)
end
# A list of lineage groups shared with your Amazon Web Services account.
# For more information, see [ Cross-Account Lineage Tracking ][1] in the
# *Amazon SageMaker Developer Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/xaccount-lineage-tracking.html
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A timestamp to filter against lineage groups created after a certain
# point in time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A timestamp to filter against lineage groups created before a certain
# point in time.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for the results. The default is `Ascending`.
#
# @option params [String] :next_token
# If the response is truncated, SageMaker returns this token. To
# retrieve the next set of algorithms, use it in the subsequent request.
#
# @option params [Integer] :max_results
# The maximum number of endpoints to return in the response. This value
# defaults to 10.
#
# @return [Types::ListLineageGroupsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListLineageGroupsResponse#lineage_group_summaries #lineage_group_summaries} => Array<Types::LineageGroupSummary>
# * {Types::ListLineageGroupsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_lineage_groups({
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.lineage_group_summaries #=> Array
# resp.lineage_group_summaries[0].lineage_group_arn #=> String
# resp.lineage_group_summaries[0].lineage_group_name #=> String
# resp.lineage_group_summaries[0].display_name #=> String
# resp.lineage_group_summaries[0].creation_time #=> Time
# resp.lineage_group_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListLineageGroups AWS API Documentation
#
# @overload list_lineage_groups(params = {})
# @param [Hash] params ({})
def list_lineage_groups(params = {}, options = {})
req = build_request(:list_lineage_groups, params)
req.send_request(options)
end
# Lists model bias jobs definitions that satisfy various filters.
#
# @option params [String] :endpoint_name
# Name of the endpoint to monitor for model bias.
#
# @option params [String] :sort_by
# Whether to sort results by the `Name` or `CreationTime` field. The
# default is `CreationTime`.
#
# @option params [String] :sort_order
# Whether to sort the results in `Ascending` or `Descending` order. The
# default is `Descending`.
#
# @option params [String] :next_token
# The token returned if the response is truncated. To retrieve the next
# set of job executions, use it in the next request.
#
# @option params [Integer] :max_results
# The maximum number of model bias jobs to return in the response. The
# default value is 10.
#
# @option params [String] :name_contains
# Filter for model bias jobs whose name contains a specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only model bias jobs created before a specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only model bias jobs created after a specified
# time.
#
# @return [Types::ListModelBiasJobDefinitionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelBiasJobDefinitionsResponse#job_definition_summaries #job_definition_summaries} => Array<Types::MonitoringJobDefinitionSummary>
# * {Types::ListModelBiasJobDefinitionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_bias_job_definitions({
# endpoint_name: "EndpointName",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# name_contains: "NameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.job_definition_summaries #=> Array
# resp.job_definition_summaries[0].monitoring_job_definition_name #=> String
# resp.job_definition_summaries[0].monitoring_job_definition_arn #=> String
# resp.job_definition_summaries[0].creation_time #=> Time
# resp.job_definition_summaries[0].endpoint_name #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelBiasJobDefinitions AWS API Documentation
#
# @overload list_model_bias_job_definitions(params = {})
# @param [Hash] params ({})
def list_model_bias_job_definitions(params = {}, options = {})
req = build_request(:list_model_bias_job_definitions, params)
req.send_request(options)
end
# List the export jobs for the Amazon SageMaker Model Card.
#
# @option params [required, String] :model_card_name
# List export jobs for the model card with the specified name.
#
# @option params [Integer] :model_card_version
# List export jobs for the model card with the specified version.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Only list model card export jobs that were created after the time
# specified.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Only list model card export jobs that were created before the time
# specified.
#
# @option params [String] :model_card_export_job_name_contains
# Only list model card export jobs with names that contain the specified
# string.
#
# @option params [String] :status_equals
# Only list model card export jobs with the specified status.
#
# @option params [String] :sort_by
# Sort model card export jobs by either name or creation time. Sorts by
# creation time by default.
#
# @option params [String] :sort_order
# Sort model card export jobs by ascending or descending order.
#
# @option params [String] :next_token
# If the response to a previous `ListModelCardExportJobs` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of model card export jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of model card export jobs to list.
#
# @return [Types::ListModelCardExportJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelCardExportJobsResponse#model_card_export_job_summaries #model_card_export_job_summaries} => Array<Types::ModelCardExportJobSummary>
# * {Types::ListModelCardExportJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_card_export_jobs({
# model_card_name: "EntityName", # required
# model_card_version: 1,
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# model_card_export_job_name_contains: "EntityName",
# status_equals: "InProgress", # accepts InProgress, Completed, Failed
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.model_card_export_job_summaries #=> Array
# resp.model_card_export_job_summaries[0].model_card_export_job_name #=> String
# resp.model_card_export_job_summaries[0].model_card_export_job_arn #=> String
# resp.model_card_export_job_summaries[0].status #=> String, one of "InProgress", "Completed", "Failed"
# resp.model_card_export_job_summaries[0].model_card_name #=> String
# resp.model_card_export_job_summaries[0].model_card_version #=> Integer
# resp.model_card_export_job_summaries[0].created_at #=> Time
# resp.model_card_export_job_summaries[0].last_modified_at #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelCardExportJobs AWS API Documentation
#
# @overload list_model_card_export_jobs(params = {})
# @param [Hash] params ({})
def list_model_card_export_jobs(params = {}, options = {})
req = build_request(:list_model_card_export_jobs, params)
req.send_request(options)
end
# List existing versions of an Amazon SageMaker Model Card.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Only list model card versions that were created after the time
# specified.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Only list model card versions that were created before the time
# specified.
#
# @option params [Integer] :max_results
# The maximum number of model card versions to list.
#
# @option params [required, String] :model_card_name
# List model card versions for the model card with the specified name.
#
# @option params [String] :model_card_status
# Only list model card versions with the specified approval status.
#
# @option params [String] :next_token
# If the response to a previous `ListModelCardVersions` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of model card versions, use the token in the next request.
#
# @option params [String] :sort_by
# Sort listed model card versions by version. Sorts by version by
# default.
#
# @option params [String] :sort_order
# Sort model card versions by ascending or descending order.
#
# @return [Types::ListModelCardVersionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelCardVersionsResponse#model_card_version_summary_list #model_card_version_summary_list} => Array<Types::ModelCardVersionSummary>
# * {Types::ListModelCardVersionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_card_versions({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# max_results: 1,
# model_card_name: "EntityName", # required
# model_card_status: "Draft", # accepts Draft, PendingReview, Approved, Archived
# next_token: "NextToken",
# sort_by: "Version", # accepts Version
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.model_card_version_summary_list #=> Array
# resp.model_card_version_summary_list[0].model_card_name #=> String
# resp.model_card_version_summary_list[0].model_card_arn #=> String
# resp.model_card_version_summary_list[0].model_card_status #=> String, one of "Draft", "PendingReview", "Approved", "Archived"
# resp.model_card_version_summary_list[0].model_card_version #=> Integer
# resp.model_card_version_summary_list[0].creation_time #=> Time
# resp.model_card_version_summary_list[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelCardVersions AWS API Documentation
#
# @overload list_model_card_versions(params = {})
# @param [Hash] params ({})
def list_model_card_versions(params = {}, options = {})
req = build_request(:list_model_card_versions, params)
req.send_request(options)
end
# List existing model cards.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# Only list model cards that were created after the time specified.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# Only list model cards that were created before the time specified.
#
# @option params [Integer] :max_results
# The maximum number of model cards to list.
#
# @option params [String] :name_contains
# Only list model cards with names that contain the specified string.
#
# @option params [String] :model_card_status
# Only list model cards with the specified approval status.
#
# @option params [String] :next_token
# If the response to a previous `ListModelCards` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of model
# cards, use the token in the next request.
#
# @option params [String] :sort_by
# Sort model cards by either name or creation time. Sorts by creation
# time by default.
#
# @option params [String] :sort_order
# Sort model cards by ascending or descending order.
#
# @return [Types::ListModelCardsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelCardsResponse#model_card_summaries #model_card_summaries} => Array<Types::ModelCardSummary>
# * {Types::ListModelCardsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_cards({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# max_results: 1,
# name_contains: "EntityName",
# model_card_status: "Draft", # accepts Draft, PendingReview, Approved, Archived
# next_token: "NextToken",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.model_card_summaries #=> Array
# resp.model_card_summaries[0].model_card_name #=> String
# resp.model_card_summaries[0].model_card_arn #=> String
# resp.model_card_summaries[0].model_card_status #=> String, one of "Draft", "PendingReview", "Approved", "Archived"
# resp.model_card_summaries[0].creation_time #=> Time
# resp.model_card_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelCards AWS API Documentation
#
# @overload list_model_cards(params = {})
# @param [Hash] params ({})
def list_model_cards(params = {}, options = {})
req = build_request(:list_model_cards, params)
req.send_request(options)
end
# Lists model explainability job definitions that satisfy various
# filters.
#
# @option params [String] :endpoint_name
# Name of the endpoint to monitor for model explainability.
#
# @option params [String] :sort_by
# Whether to sort results by the `Name` or `CreationTime` field. The
# default is `CreationTime`.
#
# @option params [String] :sort_order
# Whether to sort the results in `Ascending` or `Descending` order. The
# default is `Descending`.
#
# @option params [String] :next_token
# The token returned if the response is truncated. To retrieve the next
# set of job executions, use it in the next request.
#
# @option params [Integer] :max_results
# The maximum number of jobs to return in the response. The default
# value is 10.
#
# @option params [String] :name_contains
# Filter for model explainability jobs whose name contains a specified
# string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only model explainability jobs created before a
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only model explainability jobs created after a
# specified time.
#
# @return [Types::ListModelExplainabilityJobDefinitionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelExplainabilityJobDefinitionsResponse#job_definition_summaries #job_definition_summaries} => Array<Types::MonitoringJobDefinitionSummary>
# * {Types::ListModelExplainabilityJobDefinitionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_explainability_job_definitions({
# endpoint_name: "EndpointName",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# name_contains: "NameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.job_definition_summaries #=> Array
# resp.job_definition_summaries[0].monitoring_job_definition_name #=> String
# resp.job_definition_summaries[0].monitoring_job_definition_arn #=> String
# resp.job_definition_summaries[0].creation_time #=> Time
# resp.job_definition_summaries[0].endpoint_name #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelExplainabilityJobDefinitions AWS API Documentation
#
# @overload list_model_explainability_job_definitions(params = {})
# @param [Hash] params ({})
def list_model_explainability_job_definitions(params = {}, options = {})
req = build_request(:list_model_explainability_job_definitions, params)
req.send_request(options)
end
# Lists the domain, framework, task, and model name of standard machine
# learning models found in common model zoos.
#
# @option params [Types::ModelMetadataSearchExpression] :search_expression
# One or more filters that searches for the specified resource or
# resources in a search. All resource objects that satisfy the
# expression's condition are included in the search results. Specify
# the Framework, FrameworkVersion, Domain or Task to filter supported.
# Filter names and values are case-sensitive.
#
# @option params [String] :next_token
# If the response to a previous `ListModelMetadataResponse` request was
# truncated, the response includes a NextToken. To retrieve the next set
# of model metadata, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of models to return in the response.
#
# @return [Types::ListModelMetadataResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelMetadataResponse#model_metadata_summaries #model_metadata_summaries} => Array<Types::ModelMetadataSummary>
# * {Types::ListModelMetadataResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_metadata({
# search_expression: {
# filters: [
# {
# name: "Domain", # required, accepts Domain, Framework, Task, FrameworkVersion
# value: "String256", # required
# },
# ],
# },
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.model_metadata_summaries #=> Array
# resp.model_metadata_summaries[0].domain #=> String
# resp.model_metadata_summaries[0].framework #=> String
# resp.model_metadata_summaries[0].task #=> String
# resp.model_metadata_summaries[0].model #=> String
# resp.model_metadata_summaries[0].framework_version #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelMetadata AWS API Documentation
#
# @overload list_model_metadata(params = {})
# @param [Hash] params ({})
def list_model_metadata(params = {}, options = {})
req = build_request(:list_model_metadata, params)
req.send_request(options)
end
# Gets a list of the model groups in your Amazon Web Services account.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only model groups created after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only model groups created before the specified
# time.
#
# @option params [Integer] :max_results
# The maximum number of results to return in the response.
#
# @option params [String] :name_contains
# A string in the model group name. This filter returns only model
# groups whose name contains the specified string.
#
# @option params [String] :next_token
# If the result of the previous `ListModelPackageGroups` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of model groups, use the token in the next request.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @return [Types::ListModelPackageGroupsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelPackageGroupsOutput#model_package_group_summary_list #model_package_group_summary_list} => Array<Types::ModelPackageGroupSummary>
# * {Types::ListModelPackageGroupsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_package_groups({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# max_results: 1,
# name_contains: "NameContains",
# next_token: "NextToken",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.model_package_group_summary_list #=> Array
# resp.model_package_group_summary_list[0].model_package_group_name #=> String
# resp.model_package_group_summary_list[0].model_package_group_arn #=> String
# resp.model_package_group_summary_list[0].model_package_group_description #=> String
# resp.model_package_group_summary_list[0].creation_time #=> Time
# resp.model_package_group_summary_list[0].model_package_group_status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Deleting", "DeleteFailed"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelPackageGroups AWS API Documentation
#
# @overload list_model_package_groups(params = {})
# @param [Hash] params ({})
def list_model_package_groups(params = {}, options = {})
req = build_request(:list_model_package_groups, params)
req.send_request(options)
end
# Lists the model packages that have been created.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only model packages created after the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only model packages created before the specified
# time (timestamp).
#
# @option params [Integer] :max_results
# The maximum number of model packages to return in the response.
#
# @option params [String] :name_contains
# A string in the model package name. This filter returns only model
# packages whose name contains the specified string.
#
# @option params [String] :model_approval_status
# A filter that returns only the model packages with the specified
# approval status.
#
# @option params [String] :model_package_group_name
# A filter that returns only model versions that belong to the specified
# model group.
#
# @option params [String] :model_package_type
# A filter that returns only the model packages of the specified type.
# This can be one of the following values.
#
# * `UNVERSIONED` - List only unversioined models. This is the default
# value if no `ModelPackageType` is specified.
#
# * `VERSIONED` - List only versioned models.
#
# * `BOTH` - List both versioned and unversioned models.
#
# @option params [String] :next_token
# If the response to a previous `ListModelPackages` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of model packages, use the token in the next request.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for the results. The default is `Ascending`.
#
# @return [Types::ListModelPackagesOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelPackagesOutput#model_package_summary_list #model_package_summary_list} => Array<Types::ModelPackageSummary>
# * {Types::ListModelPackagesOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_packages({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# max_results: 1,
# name_contains: "NameContains",
# model_approval_status: "Approved", # accepts Approved, Rejected, PendingManualApproval
# model_package_group_name: "ArnOrName",
# model_package_type: "Versioned", # accepts Versioned, Unversioned, Both
# next_token: "NextToken",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.model_package_summary_list #=> Array
# resp.model_package_summary_list[0].model_package_name #=> String
# resp.model_package_summary_list[0].model_package_group_name #=> String
# resp.model_package_summary_list[0].model_package_version #=> Integer
# resp.model_package_summary_list[0].model_package_arn #=> String
# resp.model_package_summary_list[0].model_package_description #=> String
# resp.model_package_summary_list[0].creation_time #=> Time
# resp.model_package_summary_list[0].model_package_status #=> String, one of "Pending", "InProgress", "Completed", "Failed", "Deleting"
# resp.model_package_summary_list[0].model_approval_status #=> String, one of "Approved", "Rejected", "PendingManualApproval"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelPackages AWS API Documentation
#
# @overload list_model_packages(params = {})
# @param [Hash] params ({})
def list_model_packages(params = {}, options = {})
req = build_request(:list_model_packages, params)
req.send_request(options)
end
# Gets a list of model quality monitoring job definitions in your
# account.
#
# @option params [String] :endpoint_name
# A filter that returns only model quality monitoring job definitions
# that are associated with the specified endpoint.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Descending`.
#
# @option params [String] :next_token
# If the result of the previous `ListModelQualityJobDefinitions` request
# was truncated, the response includes a `NextToken`. To retrieve the
# next set of model quality monitoring job definitions, use the token in
# the next request.
#
# @option params [Integer] :max_results
# The maximum number of results to return in a call to
# `ListModelQualityJobDefinitions`.
#
# @option params [String] :name_contains
# A string in the transform job name. This filter returns only model
# quality monitoring job definitions whose name contains the specified
# string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only model quality monitoring job definitions
# created before the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only model quality monitoring job definitions
# created after the specified time.
#
# @return [Types::ListModelQualityJobDefinitionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelQualityJobDefinitionsResponse#job_definition_summaries #job_definition_summaries} => Array<Types::MonitoringJobDefinitionSummary>
# * {Types::ListModelQualityJobDefinitionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_model_quality_job_definitions({
# endpoint_name: "EndpointName",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# name_contains: "NameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.job_definition_summaries #=> Array
# resp.job_definition_summaries[0].monitoring_job_definition_name #=> String
# resp.job_definition_summaries[0].monitoring_job_definition_arn #=> String
# resp.job_definition_summaries[0].creation_time #=> Time
# resp.job_definition_summaries[0].endpoint_name #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModelQualityJobDefinitions AWS API Documentation
#
# @overload list_model_quality_job_definitions(params = {})
# @param [Hash] params ({})
def list_model_quality_job_definitions(params = {}, options = {})
req = build_request(:list_model_quality_job_definitions, params)
req.send_request(options)
end
# Lists models created with the `CreateModel` API.
#
# @option params [String] :sort_by
# Sorts the list of results. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Descending`.
#
# @option params [String] :next_token
# If the response to a previous `ListModels` request was truncated, the
# response includes a `NextToken`. To retrieve the next set of models,
# use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of models to return in the response.
#
# @option params [String] :name_contains
# A string in the model name. This filter returns only models whose name
# contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only models created before the specified time
# (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only models with a creation time greater than or
# equal to the specified time (timestamp).
#
# @return [Types::ListModelsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListModelsOutput#models #models} => Array<Types::ModelSummary>
# * {Types::ListModelsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_models({
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "PaginationToken",
# max_results: 1,
# name_contains: "ModelNameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.models #=> Array
# resp.models[0].model_name #=> String
# resp.models[0].model_arn #=> String
# resp.models[0].creation_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListModels AWS API Documentation
#
# @overload list_models(params = {})
# @param [Hash] params ({})
def list_models(params = {}, options = {})
req = build_request(:list_models, params)
req.send_request(options)
end
# Gets a list of past alerts in a model monitoring schedule.
#
# @option params [String] :monitoring_schedule_name
# The name of a monitoring schedule.
#
# @option params [String] :monitoring_alert_name
# The name of a monitoring alert.
#
# @option params [String] :sort_by
# The field used to sort results. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order, whether `Ascending` or `Descending`, of the alert
# history. The default is `Descending`.
#
# @option params [String] :next_token
# If the result of the previous `ListMonitoringAlertHistory` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of alerts in the history, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of results to display. The default is 100.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only alerts created on or before the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only alerts created on or after the specified
# time.
#
# @option params [String] :status_equals
# A filter that retrieves only alerts with a specific status.
#
# @return [Types::ListMonitoringAlertHistoryResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListMonitoringAlertHistoryResponse#monitoring_alert_history #monitoring_alert_history} => Array<Types::MonitoringAlertHistorySummary>
# * {Types::ListMonitoringAlertHistoryResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_monitoring_alert_history({
# monitoring_schedule_name: "MonitoringScheduleName",
# monitoring_alert_name: "MonitoringAlertName",
# sort_by: "CreationTime", # accepts CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# status_equals: "InAlert", # accepts InAlert, OK
# })
#
# @example Response structure
#
# resp.monitoring_alert_history #=> Array
# resp.monitoring_alert_history[0].monitoring_schedule_name #=> String
# resp.monitoring_alert_history[0].monitoring_alert_name #=> String
# resp.monitoring_alert_history[0].creation_time #=> Time
# resp.monitoring_alert_history[0].alert_status #=> String, one of "InAlert", "OK"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListMonitoringAlertHistory AWS API Documentation
#
# @overload list_monitoring_alert_history(params = {})
# @param [Hash] params ({})
def list_monitoring_alert_history(params = {}, options = {})
req = build_request(:list_monitoring_alert_history, params)
req.send_request(options)
end
# Gets the alerts for a single monitoring schedule.
#
# @option params [required, String] :monitoring_schedule_name
# The name of a monitoring schedule.
#
# @option params [String] :next_token
# If the result of the previous `ListMonitoringAlerts` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of alerts in the history, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of results to display. The default is 100.
#
# @return [Types::ListMonitoringAlertsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListMonitoringAlertsResponse#monitoring_alert_summaries #monitoring_alert_summaries} => Array<Types::MonitoringAlertSummary>
# * {Types::ListMonitoringAlertsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_monitoring_alerts({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.monitoring_alert_summaries #=> Array
# resp.monitoring_alert_summaries[0].monitoring_alert_name #=> String
# resp.monitoring_alert_summaries[0].creation_time #=> Time
# resp.monitoring_alert_summaries[0].last_modified_time #=> Time
# resp.monitoring_alert_summaries[0].alert_status #=> String, one of "InAlert", "OK"
# resp.monitoring_alert_summaries[0].datapoints_to_alert #=> Integer
# resp.monitoring_alert_summaries[0].evaluation_period #=> Integer
# resp.monitoring_alert_summaries[0].actions.model_dashboard_indicator.enabled #=> Boolean
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListMonitoringAlerts AWS API Documentation
#
# @overload list_monitoring_alerts(params = {})
# @param [Hash] params ({})
def list_monitoring_alerts(params = {}, options = {})
req = build_request(:list_monitoring_alerts, params)
req.send_request(options)
end
# Returns list of all monitoring job executions.
#
# @option params [String] :monitoring_schedule_name
# Name of a specific schedule to fetch jobs for.
#
# @option params [String] :endpoint_name
# Name of a specific endpoint to fetch jobs for.
#
# @option params [String] :sort_by
# Whether to sort results by `Status`, `CreationTime`, `ScheduledTime`
# field. The default is `CreationTime`.
#
# @option params [String] :sort_order
# Whether to sort the results in `Ascending` or `Descending` order. The
# default is `Descending`.
#
# @option params [String] :next_token
# The token returned if the response is truncated. To retrieve the next
# set of job executions, use it in the next request.
#
# @option params [Integer] :max_results
# The maximum number of jobs to return in the response. The default
# value is 10.
#
# @option params [Time,DateTime,Date,Integer,String] :scheduled_time_before
# Filter for jobs scheduled before a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :scheduled_time_after
# Filter for jobs scheduled after a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only jobs created before a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only jobs created after a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only jobs modified after a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only jobs modified before a specified time.
#
# @option params [String] :status_equals
# A filter that retrieves only jobs with a specific status.
#
# @option params [String] :monitoring_job_definition_name
# Gets a list of the monitoring job runs of the specified monitoring job
# definitions.
#
# @option params [String] :monitoring_type_equals
# A filter that returns only the monitoring job runs of the specified
# monitoring type.
#
# @return [Types::ListMonitoringExecutionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListMonitoringExecutionsResponse#monitoring_execution_summaries #monitoring_execution_summaries} => Array<Types::MonitoringExecutionSummary>
# * {Types::ListMonitoringExecutionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_monitoring_executions({
# monitoring_schedule_name: "MonitoringScheduleName",
# endpoint_name: "EndpointName",
# sort_by: "CreationTime", # accepts CreationTime, ScheduledTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# scheduled_time_before: Time.now,
# scheduled_time_after: Time.now,
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# last_modified_time_before: Time.now,
# last_modified_time_after: Time.now,
# status_equals: "Pending", # accepts Pending, Completed, CompletedWithViolations, InProgress, Failed, Stopping, Stopped
# monitoring_job_definition_name: "MonitoringJobDefinitionName",
# monitoring_type_equals: "DataQuality", # accepts DataQuality, ModelQuality, ModelBias, ModelExplainability
# })
#
# @example Response structure
#
# resp.monitoring_execution_summaries #=> Array
# resp.monitoring_execution_summaries[0].monitoring_schedule_name #=> String
# resp.monitoring_execution_summaries[0].scheduled_time #=> Time
# resp.monitoring_execution_summaries[0].creation_time #=> Time
# resp.monitoring_execution_summaries[0].last_modified_time #=> Time
# resp.monitoring_execution_summaries[0].monitoring_execution_status #=> String, one of "Pending", "Completed", "CompletedWithViolations", "InProgress", "Failed", "Stopping", "Stopped"
# resp.monitoring_execution_summaries[0].processing_job_arn #=> String
# resp.monitoring_execution_summaries[0].endpoint_name #=> String
# resp.monitoring_execution_summaries[0].failure_reason #=> String
# resp.monitoring_execution_summaries[0].monitoring_job_definition_name #=> String
# resp.monitoring_execution_summaries[0].monitoring_type #=> String, one of "DataQuality", "ModelQuality", "ModelBias", "ModelExplainability"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListMonitoringExecutions AWS API Documentation
#
# @overload list_monitoring_executions(params = {})
# @param [Hash] params ({})
def list_monitoring_executions(params = {}, options = {})
req = build_request(:list_monitoring_executions, params)
req.send_request(options)
end
# Returns list of all monitoring schedules.
#
# @option params [String] :endpoint_name
# Name of a specific endpoint to fetch schedules for.
#
# @option params [String] :sort_by
# Whether to sort results by `Status`, `CreationTime`, `ScheduledTime`
# field. The default is `CreationTime`.
#
# @option params [String] :sort_order
# Whether to sort the results in `Ascending` or `Descending` order. The
# default is `Descending`.
#
# @option params [String] :next_token
# The token returned if the response is truncated. To retrieve the next
# set of job executions, use it in the next request.
#
# @option params [Integer] :max_results
# The maximum number of jobs to return in the response. The default
# value is 10.
#
# @option params [String] :name_contains
# Filter for monitoring schedules whose name contains a specified
# string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only monitoring schedules created before a
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only monitoring schedules created after a
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only monitoring schedules modified before a
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only monitoring schedules modified after a
# specified time.
#
# @option params [String] :status_equals
# A filter that returns only monitoring schedules modified before a
# specified time.
#
# @option params [String] :monitoring_job_definition_name
# Gets a list of the monitoring schedules for the specified monitoring
# job definition.
#
# @option params [String] :monitoring_type_equals
# A filter that returns only the monitoring schedules for the specified
# monitoring type.
#
# @return [Types::ListMonitoringSchedulesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListMonitoringSchedulesResponse#monitoring_schedule_summaries #monitoring_schedule_summaries} => Array<Types::MonitoringScheduleSummary>
# * {Types::ListMonitoringSchedulesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_monitoring_schedules({
# endpoint_name: "EndpointName",
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# name_contains: "NameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# last_modified_time_before: Time.now,
# last_modified_time_after: Time.now,
# status_equals: "Pending", # accepts Pending, Failed, Scheduled, Stopped
# monitoring_job_definition_name: "MonitoringJobDefinitionName",
# monitoring_type_equals: "DataQuality", # accepts DataQuality, ModelQuality, ModelBias, ModelExplainability
# })
#
# @example Response structure
#
# resp.monitoring_schedule_summaries #=> Array
# resp.monitoring_schedule_summaries[0].monitoring_schedule_name #=> String
# resp.monitoring_schedule_summaries[0].monitoring_schedule_arn #=> String
# resp.monitoring_schedule_summaries[0].creation_time #=> Time
# resp.monitoring_schedule_summaries[0].last_modified_time #=> Time
# resp.monitoring_schedule_summaries[0].monitoring_schedule_status #=> String, one of "Pending", "Failed", "Scheduled", "Stopped"
# resp.monitoring_schedule_summaries[0].endpoint_name #=> String
# resp.monitoring_schedule_summaries[0].monitoring_job_definition_name #=> String
# resp.monitoring_schedule_summaries[0].monitoring_type #=> String, one of "DataQuality", "ModelQuality", "ModelBias", "ModelExplainability"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListMonitoringSchedules AWS API Documentation
#
# @overload list_monitoring_schedules(params = {})
# @param [Hash] params ({})
def list_monitoring_schedules(params = {}, options = {})
req = build_request(:list_monitoring_schedules, params)
req.send_request(options)
end
# Lists notebook instance lifestyle configurations created with the
# [CreateNotebookInstanceLifecycleConfig][1] API.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateNotebookInstanceLifecycleConfig.html
#
# @option params [String] :next_token
# If the result of a `ListNotebookInstanceLifecycleConfigs` request was
# truncated, the response includes a `NextToken`. To get the next set of
# lifecycle configurations, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of lifecycle configurations to return in the
# response.
#
# @option params [String] :sort_by
# Sorts the list of results. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results.
#
# @option params [String] :name_contains
# A string in the lifecycle configuration name. This filter returns only
# lifecycle configurations whose name contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only lifecycle configurations that were created
# before the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only lifecycle configurations that were created
# after the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only lifecycle configurations that were modified
# before the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only lifecycle configurations that were modified
# after the specified time (timestamp).
#
# @return [Types::ListNotebookInstanceLifecycleConfigsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListNotebookInstanceLifecycleConfigsOutput#next_token #next_token} => String
# * {Types::ListNotebookInstanceLifecycleConfigsOutput#notebook_instance_lifecycle_configs #notebook_instance_lifecycle_configs} => Array<Types::NotebookInstanceLifecycleConfigSummary>
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_notebook_instance_lifecycle_configs({
# next_token: "NextToken",
# max_results: 1,
# sort_by: "Name", # accepts Name, CreationTime, LastModifiedTime
# sort_order: "Ascending", # accepts Ascending, Descending
# name_contains: "NotebookInstanceLifecycleConfigNameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# last_modified_time_before: Time.now,
# last_modified_time_after: Time.now,
# })
#
# @example Response structure
#
# resp.next_token #=> String
# resp.notebook_instance_lifecycle_configs #=> Array
# resp.notebook_instance_lifecycle_configs[0].notebook_instance_lifecycle_config_name #=> String
# resp.notebook_instance_lifecycle_configs[0].notebook_instance_lifecycle_config_arn #=> String
# resp.notebook_instance_lifecycle_configs[0].creation_time #=> Time
# resp.notebook_instance_lifecycle_configs[0].last_modified_time #=> Time
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListNotebookInstanceLifecycleConfigs AWS API Documentation
#
# @overload list_notebook_instance_lifecycle_configs(params = {})
# @param [Hash] params ({})
def list_notebook_instance_lifecycle_configs(params = {}, options = {})
req = build_request(:list_notebook_instance_lifecycle_configs, params)
req.send_request(options)
end
# Returns a list of the SageMaker notebook instances in the requester's
# account in an Amazon Web Services Region.
#
# @option params [String] :next_token
# If the previous call to the `ListNotebookInstances` is truncated, the
# response includes a `NextToken`. You can use this token in your
# subsequent `ListNotebookInstances` request to fetch the next set of
# notebook instances.
#
# You might specify a filter or a sort order in your request. When
# response is truncated, you must use the same values for the filer and
# sort order in the next request.
#
#
#
# @option params [Integer] :max_results
# The maximum number of notebook instances to return.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `Name`.
#
# @option params [String] :sort_order
# The sort order for results.
#
# @option params [String] :name_contains
# A string in the notebook instances' name. This filter returns only
# notebook instances whose name contains the specified string.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only notebook instances that were created before
# the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only notebook instances that were created after
# the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only notebook instances that were modified
# before the specified time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only notebook instances that were modified after
# the specified time (timestamp).
#
# @option params [String] :status_equals
# A filter that returns only notebook instances with the specified
# status.
#
# @option params [String] :notebook_instance_lifecycle_config_name_contains
# A string in the name of a notebook instances lifecycle configuration
# associated with this notebook instance. This filter returns only
# notebook instances associated with a lifecycle configuration with a
# name that contains the specified string.
#
# @option params [String] :default_code_repository_contains
# A string in the name or URL of a Git repository associated with this
# notebook instance. This filter returns only notebook instances
# associated with a git repository with a name that contains the
# specified string.
#
# @option params [String] :additional_code_repository_equals
# A filter that returns only notebook instances with associated with the
# specified git repository.
#
# @return [Types::ListNotebookInstancesOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListNotebookInstancesOutput#next_token #next_token} => String
# * {Types::ListNotebookInstancesOutput#notebook_instances #notebook_instances} => Array<Types::NotebookInstanceSummary>
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_notebook_instances({
# next_token: "NextToken",
# max_results: 1,
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# name_contains: "NotebookInstanceNameContains",
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# last_modified_time_before: Time.now,
# last_modified_time_after: Time.now,
# status_equals: "Pending", # accepts Pending, InService, Stopping, Stopped, Failed, Deleting, Updating
# notebook_instance_lifecycle_config_name_contains: "NotebookInstanceLifecycleConfigName",
# default_code_repository_contains: "CodeRepositoryContains",
# additional_code_repository_equals: "CodeRepositoryNameOrUrl",
# })
#
# @example Response structure
#
# resp.next_token #=> String
# resp.notebook_instances #=> Array
# resp.notebook_instances[0].notebook_instance_name #=> String
# resp.notebook_instances[0].notebook_instance_arn #=> String
# resp.notebook_instances[0].notebook_instance_status #=> String, one of "Pending", "InService", "Stopping", "Stopped", "Failed", "Deleting", "Updating"
# resp.notebook_instances[0].url #=> String
# resp.notebook_instances[0].instance_type #=> String, one of "ml.t2.medium", "ml.t2.large", "ml.t2.xlarge", "ml.t2.2xlarge", "ml.t3.medium", "ml.t3.large", "ml.t3.xlarge", "ml.t3.2xlarge", "ml.m4.xlarge", "ml.m4.2xlarge", "ml.m4.4xlarge", "ml.m4.10xlarge", "ml.m4.16xlarge", "ml.m5.xlarge", "ml.m5.2xlarge", "ml.m5.4xlarge", "ml.m5.12xlarge", "ml.m5.24xlarge", "ml.m5d.large", "ml.m5d.xlarge", "ml.m5d.2xlarge", "ml.m5d.4xlarge", "ml.m5d.8xlarge", "ml.m5d.12xlarge", "ml.m5d.16xlarge", "ml.m5d.24xlarge", "ml.c4.xlarge", "ml.c4.2xlarge", "ml.c4.4xlarge", "ml.c4.8xlarge", "ml.c5.xlarge", "ml.c5.2xlarge", "ml.c5.4xlarge", "ml.c5.9xlarge", "ml.c5.18xlarge", "ml.c5d.xlarge", "ml.c5d.2xlarge", "ml.c5d.4xlarge", "ml.c5d.9xlarge", "ml.c5d.18xlarge", "ml.p2.xlarge", "ml.p2.8xlarge", "ml.p2.16xlarge", "ml.p3.2xlarge", "ml.p3.8xlarge", "ml.p3.16xlarge", "ml.p3dn.24xlarge", "ml.g4dn.xlarge", "ml.g4dn.2xlarge", "ml.g4dn.4xlarge", "ml.g4dn.8xlarge", "ml.g4dn.12xlarge", "ml.g4dn.16xlarge", "ml.r5.large", "ml.r5.xlarge", "ml.r5.2xlarge", "ml.r5.4xlarge", "ml.r5.8xlarge", "ml.r5.12xlarge", "ml.r5.16xlarge", "ml.r5.24xlarge", "ml.g5.xlarge", "ml.g5.2xlarge", "ml.g5.4xlarge", "ml.g5.8xlarge", "ml.g5.16xlarge", "ml.g5.12xlarge", "ml.g5.24xlarge", "ml.g5.48xlarge"
# resp.notebook_instances[0].creation_time #=> Time
# resp.notebook_instances[0].last_modified_time #=> Time
# resp.notebook_instances[0].notebook_instance_lifecycle_config_name #=> String
# resp.notebook_instances[0].default_code_repository #=> String
# resp.notebook_instances[0].additional_code_repositories #=> Array
# resp.notebook_instances[0].additional_code_repositories[0] #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListNotebookInstances AWS API Documentation
#
# @overload list_notebook_instances(params = {})
# @param [Hash] params ({})
def list_notebook_instances(params = {}, options = {})
req = build_request(:list_notebook_instances, params)
req.send_request(options)
end
# Gets a list of `PipeLineExecutionStep` objects.
#
# @option params [String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @option params [String] :next_token
# If the result of the previous `ListPipelineExecutionSteps` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of pipeline execution steps, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of pipeline execution steps to return in the
# response.
#
# @option params [String] :sort_order
# The field by which to sort results. The default is `CreatedTime`.
#
# @return [Types::ListPipelineExecutionStepsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListPipelineExecutionStepsResponse#pipeline_execution_steps #pipeline_execution_steps} => Array<Types::PipelineExecutionStep>
# * {Types::ListPipelineExecutionStepsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_pipeline_execution_steps({
# pipeline_execution_arn: "PipelineExecutionArn",
# next_token: "NextToken",
# max_results: 1,
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.pipeline_execution_steps #=> Array
# resp.pipeline_execution_steps[0].step_name #=> String
# resp.pipeline_execution_steps[0].step_display_name #=> String
# resp.pipeline_execution_steps[0].step_description #=> String
# resp.pipeline_execution_steps[0].start_time #=> Time
# resp.pipeline_execution_steps[0].end_time #=> Time
# resp.pipeline_execution_steps[0].step_status #=> String, one of "Starting", "Executing", "Stopping", "Stopped", "Failed", "Succeeded"
# resp.pipeline_execution_steps[0].cache_hit_result.source_pipeline_execution_arn #=> String
# resp.pipeline_execution_steps[0].attempt_count #=> Integer
# resp.pipeline_execution_steps[0].failure_reason #=> String
# resp.pipeline_execution_steps[0].metadata.training_job.arn #=> String
# resp.pipeline_execution_steps[0].metadata.processing_job.arn #=> String
# resp.pipeline_execution_steps[0].metadata.transform_job.arn #=> String
# resp.pipeline_execution_steps[0].metadata.tuning_job.arn #=> String
# resp.pipeline_execution_steps[0].metadata.model.arn #=> String
# resp.pipeline_execution_steps[0].metadata.register_model.arn #=> String
# resp.pipeline_execution_steps[0].metadata.condition.outcome #=> String, one of "True", "False"
# resp.pipeline_execution_steps[0].metadata.callback.callback_token #=> String
# resp.pipeline_execution_steps[0].metadata.callback.sqs_queue_url #=> String
# resp.pipeline_execution_steps[0].metadata.callback.output_parameters #=> Array
# resp.pipeline_execution_steps[0].metadata.callback.output_parameters[0].name #=> String
# resp.pipeline_execution_steps[0].metadata.callback.output_parameters[0].value #=> String
# resp.pipeline_execution_steps[0].metadata.lambda.arn #=> String
# resp.pipeline_execution_steps[0].metadata.lambda.output_parameters #=> Array
# resp.pipeline_execution_steps[0].metadata.lambda.output_parameters[0].name #=> String
# resp.pipeline_execution_steps[0].metadata.lambda.output_parameters[0].value #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.check_type #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.baseline_used_for_drift_check_statistics #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.baseline_used_for_drift_check_constraints #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.calculated_baseline_statistics #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.calculated_baseline_constraints #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.model_package_group_name #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.violation_report #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.check_job_arn #=> String
# resp.pipeline_execution_steps[0].metadata.quality_check.skip_check #=> Boolean
# resp.pipeline_execution_steps[0].metadata.quality_check.register_new_baseline #=> Boolean
# resp.pipeline_execution_steps[0].metadata.clarify_check.check_type #=> String
# resp.pipeline_execution_steps[0].metadata.clarify_check.baseline_used_for_drift_check_constraints #=> String
# resp.pipeline_execution_steps[0].metadata.clarify_check.calculated_baseline_constraints #=> String
# resp.pipeline_execution_steps[0].metadata.clarify_check.model_package_group_name #=> String
# resp.pipeline_execution_steps[0].metadata.clarify_check.violation_report #=> String
# resp.pipeline_execution_steps[0].metadata.clarify_check.check_job_arn #=> String
# resp.pipeline_execution_steps[0].metadata.clarify_check.skip_check #=> Boolean
# resp.pipeline_execution_steps[0].metadata.clarify_check.register_new_baseline #=> Boolean
# resp.pipeline_execution_steps[0].metadata.emr.cluster_id #=> String
# resp.pipeline_execution_steps[0].metadata.emr.step_id #=> String
# resp.pipeline_execution_steps[0].metadata.emr.step_name #=> String
# resp.pipeline_execution_steps[0].metadata.emr.log_file_path #=> String
# resp.pipeline_execution_steps[0].metadata.fail.error_message #=> String
# resp.pipeline_execution_steps[0].metadata.auto_ml_job.arn #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListPipelineExecutionSteps AWS API Documentation
#
# @overload list_pipeline_execution_steps(params = {})
# @param [Hash] params ({})
def list_pipeline_execution_steps(params = {}, options = {})
req = build_request(:list_pipeline_execution_steps, params)
req.send_request(options)
end
# Gets a list of the pipeline executions.
#
# @option params [required, String] :pipeline_name
# The name of the pipeline.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns the pipeline executions that were created after
# a specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns the pipeline executions that were created before
# a specified time.
#
# @option params [String] :sort_by
# The field by which to sort results. The default is `CreatedTime`.
#
# @option params [String] :sort_order
# The sort order for results.
#
# @option params [String] :next_token
# If the result of the previous `ListPipelineExecutions` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of pipeline executions, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of pipeline executions to return in the response.
#
# @return [Types::ListPipelineExecutionsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListPipelineExecutionsResponse#pipeline_execution_summaries #pipeline_execution_summaries} => Array<Types::PipelineExecutionSummary>
# * {Types::ListPipelineExecutionsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_pipeline_executions({
# pipeline_name: "PipelineNameOrArn", # required
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "CreationTime", # accepts CreationTime, PipelineExecutionArn
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.pipeline_execution_summaries #=> Array
# resp.pipeline_execution_summaries[0].pipeline_execution_arn #=> String
# resp.pipeline_execution_summaries[0].start_time #=> Time
# resp.pipeline_execution_summaries[0].pipeline_execution_status #=> String, one of "Executing", "Stopping", "Stopped", "Failed", "Succeeded"
# resp.pipeline_execution_summaries[0].pipeline_execution_description #=> String
# resp.pipeline_execution_summaries[0].pipeline_execution_display_name #=> String
# resp.pipeline_execution_summaries[0].pipeline_execution_failure_reason #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListPipelineExecutions AWS API Documentation
#
# @overload list_pipeline_executions(params = {})
# @param [Hash] params ({})
def list_pipeline_executions(params = {}, options = {})
req = build_request(:list_pipeline_executions, params)
req.send_request(options)
end
# Gets a list of parameters for a pipeline execution.
#
# @option params [required, String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @option params [String] :next_token
# If the result of the previous `ListPipelineParametersForExecution`
# request was truncated, the response includes a `NextToken`. To
# retrieve the next set of parameters, use the token in the next
# request.
#
# @option params [Integer] :max_results
# The maximum number of parameters to return in the response.
#
# @return [Types::ListPipelineParametersForExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListPipelineParametersForExecutionResponse#pipeline_parameters #pipeline_parameters} => Array<Types::Parameter>
# * {Types::ListPipelineParametersForExecutionResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_pipeline_parameters_for_execution({
# pipeline_execution_arn: "PipelineExecutionArn", # required
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.pipeline_parameters #=> Array
# resp.pipeline_parameters[0].name #=> String
# resp.pipeline_parameters[0].value #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListPipelineParametersForExecution AWS API Documentation
#
# @overload list_pipeline_parameters_for_execution(params = {})
# @param [Hash] params ({})
def list_pipeline_parameters_for_execution(params = {}, options = {})
req = build_request(:list_pipeline_parameters_for_execution, params)
req.send_request(options)
end
# Gets a list of pipelines.
#
# @option params [String] :pipeline_name_prefix
# The prefix of the pipeline name.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns the pipelines that were created after a
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns the pipelines that were created before a
# specified time.
#
# @option params [String] :sort_by
# The field by which to sort results. The default is `CreatedTime`.
#
# @option params [String] :sort_order
# The sort order for results.
#
# @option params [String] :next_token
# If the result of the previous `ListPipelines` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of
# pipelines, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of pipelines to return in the response.
#
# @return [Types::ListPipelinesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListPipelinesResponse#pipeline_summaries #pipeline_summaries} => Array<Types::PipelineSummary>
# * {Types::ListPipelinesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_pipelines({
# pipeline_name_prefix: "PipelineName",
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.pipeline_summaries #=> Array
# resp.pipeline_summaries[0].pipeline_arn #=> String
# resp.pipeline_summaries[0].pipeline_name #=> String
# resp.pipeline_summaries[0].pipeline_display_name #=> String
# resp.pipeline_summaries[0].pipeline_description #=> String
# resp.pipeline_summaries[0].role_arn #=> String
# resp.pipeline_summaries[0].creation_time #=> Time
# resp.pipeline_summaries[0].last_modified_time #=> Time
# resp.pipeline_summaries[0].last_execution_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListPipelines AWS API Documentation
#
# @overload list_pipelines(params = {})
# @param [Hash] params ({})
def list_pipelines(params = {}, options = {})
req = build_request(:list_pipelines, params)
req.send_request(options)
end
# Lists processing jobs that satisfy various filters.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only processing jobs created after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only processing jobs created after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only processing jobs modified after the
# specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only processing jobs modified before the
# specified time.
#
# @option params [String] :name_contains
# A string in the processing job name. This filter returns only
# processing jobs whose name contains the specified string.
#
# @option params [String] :status_equals
# A filter that retrieves only processing jobs with a specific status.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @option params [String] :next_token
# If the result of the previous `ListProcessingJobs` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of processing jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of processing jobs to return in the response.
#
# @return [Types::ListProcessingJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListProcessingJobsResponse#processing_job_summaries #processing_job_summaries} => Array<Types::ProcessingJobSummary>
# * {Types::ListProcessingJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_processing_jobs({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "String",
# status_equals: "InProgress", # accepts InProgress, Completed, Failed, Stopping, Stopped
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.processing_job_summaries #=> Array
# resp.processing_job_summaries[0].processing_job_name #=> String
# resp.processing_job_summaries[0].processing_job_arn #=> String
# resp.processing_job_summaries[0].creation_time #=> Time
# resp.processing_job_summaries[0].processing_end_time #=> Time
# resp.processing_job_summaries[0].last_modified_time #=> Time
# resp.processing_job_summaries[0].processing_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.processing_job_summaries[0].failure_reason #=> String
# resp.processing_job_summaries[0].exit_message #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListProcessingJobs AWS API Documentation
#
# @overload list_processing_jobs(params = {})
# @param [Hash] params ({})
def list_processing_jobs(params = {}, options = {})
req = build_request(:list_processing_jobs, params)
req.send_request(options)
end
# Gets a list of the projects in an Amazon Web Services account.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns the projects that were created after a specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns the projects that were created before a
# specified time.
#
# @option params [Integer] :max_results
# The maximum number of projects to return in the response.
#
# @option params [String] :name_contains
# A filter that returns the projects whose name contains a specified
# string.
#
# @option params [String] :next_token
# If the result of the previous `ListProjects` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of
# projects, use the token in the next request.
#
# @option params [String] :sort_by
# The field by which to sort results. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @return [Types::ListProjectsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListProjectsOutput#project_summary_list #project_summary_list} => Array<Types::ProjectSummary>
# * {Types::ListProjectsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_projects({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# max_results: 1,
# name_contains: "ProjectEntityName",
# next_token: "NextToken",
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.project_summary_list #=> Array
# resp.project_summary_list[0].project_name #=> String
# resp.project_summary_list[0].project_description #=> String
# resp.project_summary_list[0].project_arn #=> String
# resp.project_summary_list[0].project_id #=> String
# resp.project_summary_list[0].creation_time #=> Time
# resp.project_summary_list[0].project_status #=> String, one of "Pending", "CreateInProgress", "CreateCompleted", "CreateFailed", "DeleteInProgress", "DeleteFailed", "DeleteCompleted", "UpdateInProgress", "UpdateCompleted", "UpdateFailed"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListProjects AWS API Documentation
#
# @overload list_projects(params = {})
# @param [Hash] params ({})
def list_projects(params = {}, options = {})
req = build_request(:list_projects, params)
req.send_request(options)
end
# Lists spaces.
#
# @option params [String] :next_token
# If the previous response was truncated, you will receive this token.
# Use it in your next request to receive the next set of results.
#
# @option params [Integer] :max_results
# Returns a list up to a specified limit.
#
# @option params [String] :sort_order
# The sort order for the results. The default is `Ascending`.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# `CreationTime`.
#
# @option params [String] :domain_id_equals
# A parameter to search for the Domain ID.
#
# @option params [String] :space_name_contains
# A parameter by which to filter the results.
#
# @return [Types::ListSpacesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListSpacesResponse#spaces #spaces} => Array<Types::SpaceDetails>
# * {Types::ListSpacesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_spaces({
# next_token: "NextToken",
# max_results: 1,
# sort_order: "Ascending", # accepts Ascending, Descending
# sort_by: "CreationTime", # accepts CreationTime, LastModifiedTime
# domain_id_equals: "DomainId",
# space_name_contains: "SpaceName",
# })
#
# @example Response structure
#
# resp.spaces #=> Array
# resp.spaces[0].domain_id #=> String
# resp.spaces[0].space_name #=> String
# resp.spaces[0].status #=> String, one of "Deleting", "Failed", "InService", "Pending", "Updating", "Update_Failed", "Delete_Failed"
# resp.spaces[0].creation_time #=> Time
# resp.spaces[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListSpaces AWS API Documentation
#
# @overload list_spaces(params = {})
# @param [Hash] params ({})
def list_spaces(params = {}, options = {})
req = build_request(:list_spaces, params)
req.send_request(options)
end
# Lists devices allocated to the stage, containing detailed device
# information and deployment status.
#
# @option params [String] :next_token
# The response from the last list when returning a list large enough to
# neeed tokening.
#
# @option params [Integer] :max_results
# The maximum number of requests to select.
#
# @option params [required, String] :edge_deployment_plan_name
# The name of the edge deployment plan.
#
# @option params [Boolean] :exclude_devices_deployed_in_other_stage
# Toggle for excluding devices deployed in other stages.
#
# @option params [required, String] :stage_name
# The name of the stage in the deployment.
#
# @return [Types::ListStageDevicesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListStageDevicesResponse#device_deployment_summaries #device_deployment_summaries} => Array<Types::DeviceDeploymentSummary>
# * {Types::ListStageDevicesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_stage_devices({
# next_token: "NextToken",
# max_results: 1,
# edge_deployment_plan_name: "EntityName", # required
# exclude_devices_deployed_in_other_stage: false,
# stage_name: "EntityName", # required
# })
#
# @example Response structure
#
# resp.device_deployment_summaries #=> Array
# resp.device_deployment_summaries[0].edge_deployment_plan_arn #=> String
# resp.device_deployment_summaries[0].edge_deployment_plan_name #=> String
# resp.device_deployment_summaries[0].stage_name #=> String
# resp.device_deployment_summaries[0].deployed_stage_name #=> String
# resp.device_deployment_summaries[0].device_fleet_name #=> String
# resp.device_deployment_summaries[0].device_name #=> String
# resp.device_deployment_summaries[0].device_arn #=> String
# resp.device_deployment_summaries[0].device_deployment_status #=> String, one of "READYTODEPLOY", "INPROGRESS", "DEPLOYED", "FAILED", "STOPPING", "STOPPED"
# resp.device_deployment_summaries[0].device_deployment_status_message #=> String
# resp.device_deployment_summaries[0].description #=> String
# resp.device_deployment_summaries[0].deployment_start_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListStageDevices AWS API Documentation
#
# @overload list_stage_devices(params = {})
# @param [Hash] params ({})
def list_stage_devices(params = {}, options = {})
req = build_request(:list_stage_devices, params)
req.send_request(options)
end
# Lists the Studio Lifecycle Configurations in your Amazon Web Services
# Account.
#
# @option params [Integer] :max_results
# The maximum number of Studio Lifecycle Configurations to return in the
# response. The default value is 10.
#
# @option params [String] :next_token
# If the previous call to ListStudioLifecycleConfigs didn't return the
# full set of Lifecycle Configurations, the call returns a token for
# getting the next set of Lifecycle Configurations.
#
# @option params [String] :name_contains
# A string in the Lifecycle Configuration name. This filter returns only
# Lifecycle Configurations whose name contains the specified string.
#
# @option params [String] :app_type_equals
# A parameter to search for the App Type to which the Lifecycle
# Configuration is attached.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only Lifecycle Configurations created on or
# before the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only Lifecycle Configurations created on or
# after the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :modified_time_before
# A filter that returns only Lifecycle Configurations modified before
# the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :modified_time_after
# A filter that returns only Lifecycle Configurations modified after the
# specified time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is CreationTime.
#
# @option params [String] :sort_order
# The sort order. The default value is Descending.
#
# @return [Types::ListStudioLifecycleConfigsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListStudioLifecycleConfigsResponse#next_token #next_token} => String
# * {Types::ListStudioLifecycleConfigsResponse#studio_lifecycle_configs #studio_lifecycle_configs} => Array<Types::StudioLifecycleConfigDetails>
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_studio_lifecycle_configs({
# max_results: 1,
# next_token: "NextToken",
# name_contains: "StudioLifecycleConfigName",
# app_type_equals: "JupyterServer", # accepts JupyterServer, KernelGateway
# creation_time_before: Time.now,
# creation_time_after: Time.now,
# modified_time_before: Time.now,
# modified_time_after: Time.now,
# sort_by: "CreationTime", # accepts CreationTime, LastModifiedTime, Name
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.next_token #=> String
# resp.studio_lifecycle_configs #=> Array
# resp.studio_lifecycle_configs[0].studio_lifecycle_config_arn #=> String
# resp.studio_lifecycle_configs[0].studio_lifecycle_config_name #=> String
# resp.studio_lifecycle_configs[0].creation_time #=> Time
# resp.studio_lifecycle_configs[0].last_modified_time #=> Time
# resp.studio_lifecycle_configs[0].studio_lifecycle_config_app_type #=> String, one of "JupyterServer", "KernelGateway"
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListStudioLifecycleConfigs AWS API Documentation
#
# @overload list_studio_lifecycle_configs(params = {})
# @param [Hash] params ({})
def list_studio_lifecycle_configs(params = {}, options = {})
req = build_request(:list_studio_lifecycle_configs, params)
req.send_request(options)
end
# Gets a list of the work teams that you are subscribed to in the Amazon
# Web Services Marketplace. The list may be empty if no work team
# satisfies the filter specified in the `NameContains` parameter.
#
# @option params [String] :name_contains
# A string in the work team name. This filter returns only work teams
# whose name contains the specified string.
#
# @option params [String] :next_token
# If the result of the previous `ListSubscribedWorkteams` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of labeling jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of work teams to return in each page of the
# response.
#
# @return [Types::ListSubscribedWorkteamsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListSubscribedWorkteamsResponse#subscribed_workteams #subscribed_workteams} => Array<Types::SubscribedWorkteam>
# * {Types::ListSubscribedWorkteamsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_subscribed_workteams({
# name_contains: "WorkteamName",
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.subscribed_workteams #=> Array
# resp.subscribed_workteams[0].workteam_arn #=> String
# resp.subscribed_workteams[0].marketplace_title #=> String
# resp.subscribed_workteams[0].seller_name #=> String
# resp.subscribed_workteams[0].marketplace_description #=> String
# resp.subscribed_workteams[0].listing_id #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListSubscribedWorkteams AWS API Documentation
#
# @overload list_subscribed_workteams(params = {})
# @param [Hash] params ({})
def list_subscribed_workteams(params = {}, options = {})
req = build_request(:list_subscribed_workteams, params)
req.send_request(options)
end
# Returns the tags for the specified SageMaker resource.
#
# @option params [required, String] :resource_arn
# The Amazon Resource Name (ARN) of the resource whose tags you want to
# retrieve.
#
# @option params [String] :next_token
# If the response to the previous `ListTags` request is truncated,
# SageMaker returns this token. To retrieve the next set of tags, use it
# in the subsequent request.
#
# @option params [Integer] :max_results
# Maximum number of tags to return.
#
# @return [Types::ListTagsOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListTagsOutput#tags #tags} => Array<Types::Tag>
# * {Types::ListTagsOutput#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_tags({
# resource_arn: "ResourceArn", # required
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.tags #=> Array
# resp.tags[0].key #=> String
# resp.tags[0].value #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListTags AWS API Documentation
#
# @overload list_tags(params = {})
# @param [Hash] params ({})
def list_tags(params = {}, options = {})
req = build_request(:list_tags, params)
req.send_request(options)
end
# Lists training jobs.
#
# When `StatusEquals` and `MaxResults` are set at the same time, the
# `MaxResults` number of training jobs are first retrieved ignoring the
# `StatusEquals` parameter and then they are filtered by the
# `StatusEquals` parameter, which is returned as a response.
#
# For example, if `ListTrainingJobs` is invoked with the following
# parameters:
#
# `\{ ... MaxResults: 100, StatusEquals: InProgress ... \}`
#
# First, 100 trainings jobs with any status, including those other than
# `InProgress`, are selected (sorted according to the creation time,
# from the most current to the oldest). Next, those with a status of
# `InProgress` are returned.
#
# You can quickly test the API using the following Amazon Web Services
# CLI code.
#
# `aws sagemaker list-training-jobs --max-results 100 --status-equals
# InProgress`
#
#
#
# @option params [String] :next_token
# If the result of the previous `ListTrainingJobs` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of training jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of training jobs to return in the response.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only training jobs created after the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only training jobs created before the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only training jobs modified after the specified
# time (timestamp).
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only training jobs modified before the specified
# time (timestamp).
#
# @option params [String] :name_contains
# A string in the training job name. This filter returns only training
# jobs whose name contains the specified string.
#
# @option params [String] :status_equals
# A filter that retrieves only training jobs with a specific status.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @option params [String] :warm_pool_status_equals
# A filter that retrieves only training jobs with a specific warm pool
# status.
#
# @return [Types::ListTrainingJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListTrainingJobsResponse#training_job_summaries #training_job_summaries} => Array<Types::TrainingJobSummary>
# * {Types::ListTrainingJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_training_jobs({
# next_token: "NextToken",
# max_results: 1,
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# status_equals: "InProgress", # accepts InProgress, Completed, Failed, Stopping, Stopped
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# warm_pool_status_equals: "Available", # accepts Available, Terminated, Reused, InUse
# })
#
# @example Response structure
#
# resp.training_job_summaries #=> Array
# resp.training_job_summaries[0].training_job_name #=> String
# resp.training_job_summaries[0].training_job_arn #=> String
# resp.training_job_summaries[0].creation_time #=> Time
# resp.training_job_summaries[0].training_end_time #=> Time
# resp.training_job_summaries[0].last_modified_time #=> Time
# resp.training_job_summaries[0].training_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.training_job_summaries[0].warm_pool_status.status #=> String, one of "Available", "Terminated", "Reused", "InUse"
# resp.training_job_summaries[0].warm_pool_status.resource_retained_billable_time_in_seconds #=> Integer
# resp.training_job_summaries[0].warm_pool_status.reused_by_job #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListTrainingJobs AWS API Documentation
#
# @overload list_training_jobs(params = {})
# @param [Hash] params ({})
def list_training_jobs(params = {}, options = {})
req = build_request(:list_training_jobs, params)
req.send_request(options)
end
# Gets a list of [TrainingJobSummary][1] objects that describe the
# training jobs that a hyperparameter tuning job launched.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_TrainingJobSummary.html
#
# @option params [required, String] :hyper_parameter_tuning_job_name
# The name of the tuning job whose training jobs you want to list.
#
# @option params [String] :next_token
# If the result of the previous
# `ListTrainingJobsForHyperParameterTuningJob` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of
# training jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of training jobs to return. The default value is
# 10.
#
# @option params [String] :status_equals
# A filter that returns only training jobs with the specified status.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `Name`.
#
# If the value of this field is `FinalObjectiveMetricValue`, any
# training jobs that did not return an objective metric are not listed.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @return [Types::ListTrainingJobsForHyperParameterTuningJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListTrainingJobsForHyperParameterTuningJobResponse#training_job_summaries #training_job_summaries} => Array<Types::HyperParameterTrainingJobSummary>
# * {Types::ListTrainingJobsForHyperParameterTuningJobResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_training_jobs_for_hyper_parameter_tuning_job({
# hyper_parameter_tuning_job_name: "HyperParameterTuningJobName", # required
# next_token: "NextToken",
# max_results: 1,
# status_equals: "InProgress", # accepts InProgress, Completed, Failed, Stopping, Stopped
# sort_by: "Name", # accepts Name, CreationTime, Status, FinalObjectiveMetricValue
# sort_order: "Ascending", # accepts Ascending, Descending
# })
#
# @example Response structure
#
# resp.training_job_summaries #=> Array
# resp.training_job_summaries[0].training_job_definition_name #=> String
# resp.training_job_summaries[0].training_job_name #=> String
# resp.training_job_summaries[0].training_job_arn #=> String
# resp.training_job_summaries[0].tuning_job_name #=> String
# resp.training_job_summaries[0].creation_time #=> Time
# resp.training_job_summaries[0].training_start_time #=> Time
# resp.training_job_summaries[0].training_end_time #=> Time
# resp.training_job_summaries[0].training_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.training_job_summaries[0].tuned_hyper_parameters #=> Hash
# resp.training_job_summaries[0].tuned_hyper_parameters["HyperParameterKey"] #=> String
# resp.training_job_summaries[0].failure_reason #=> String
# resp.training_job_summaries[0].final_hyper_parameter_tuning_job_objective_metric.type #=> String, one of "Maximize", "Minimize"
# resp.training_job_summaries[0].final_hyper_parameter_tuning_job_objective_metric.metric_name #=> String
# resp.training_job_summaries[0].final_hyper_parameter_tuning_job_objective_metric.value #=> Float
# resp.training_job_summaries[0].objective_status #=> String, one of "Succeeded", "Pending", "Failed"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListTrainingJobsForHyperParameterTuningJob AWS API Documentation
#
# @overload list_training_jobs_for_hyper_parameter_tuning_job(params = {})
# @param [Hash] params ({})
def list_training_jobs_for_hyper_parameter_tuning_job(params = {}, options = {})
req = build_request(:list_training_jobs_for_hyper_parameter_tuning_job, params)
req.send_request(options)
end
# Lists transform jobs.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_after
# A filter that returns only transform jobs created after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :creation_time_before
# A filter that returns only transform jobs created before the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_after
# A filter that returns only transform jobs modified after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :last_modified_time_before
# A filter that returns only transform jobs modified before the
# specified time.
#
# @option params [String] :name_contains
# A string in the transform job name. This filter returns only transform
# jobs whose name contains the specified string.
#
# @option params [String] :status_equals
# A filter that retrieves only transform jobs with a specific status.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Descending`.
#
# @option params [String] :next_token
# If the result of the previous `ListTransformJobs` request was
# truncated, the response includes a `NextToken`. To retrieve the next
# set of transform jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of transform jobs to return in the response. The
# default value is `10`.
#
# @return [Types::ListTransformJobsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListTransformJobsResponse#transform_job_summaries #transform_job_summaries} => Array<Types::TransformJobSummary>
# * {Types::ListTransformJobsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_transform_jobs({
# creation_time_after: Time.now,
# creation_time_before: Time.now,
# last_modified_time_after: Time.now,
# last_modified_time_before: Time.now,
# name_contains: "NameContains",
# status_equals: "InProgress", # accepts InProgress, Completed, Failed, Stopping, Stopped
# sort_by: "Name", # accepts Name, CreationTime, Status
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.transform_job_summaries #=> Array
# resp.transform_job_summaries[0].transform_job_name #=> String
# resp.transform_job_summaries[0].transform_job_arn #=> String
# resp.transform_job_summaries[0].creation_time #=> Time
# resp.transform_job_summaries[0].transform_end_time #=> Time
# resp.transform_job_summaries[0].last_modified_time #=> Time
# resp.transform_job_summaries[0].transform_job_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.transform_job_summaries[0].failure_reason #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListTransformJobs AWS API Documentation
#
# @overload list_transform_jobs(params = {})
# @param [Hash] params ({})
def list_transform_jobs(params = {}, options = {})
req = build_request(:list_transform_jobs, params)
req.send_request(options)
end
# Lists the trial components in your account. You can sort the list by
# trial component name or creation time. You can filter the list to show
# only components that were created in a specific time range. You can
# also filter on one of the following:
#
# * `ExperimentName`
#
# * `SourceArn`
#
# * `TrialName`
#
# @option params [String] :experiment_name
# A filter that returns only components that are part of the specified
# experiment. If you specify `ExperimentName`, you can't filter by
# `SourceArn` or `TrialName`.
#
# @option params [String] :trial_name
# A filter that returns only components that are part of the specified
# trial. If you specify `TrialName`, you can't filter by
# `ExperimentName` or `SourceArn`.
#
# @option params [String] :source_arn
# A filter that returns only components that have the specified source
# Amazon Resource Name (ARN). If you specify `SourceArn`, you can't
# filter by `ExperimentName` or `TrialName`.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only components created after the specified
# time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only components created before the specified
# time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [Integer] :max_results
# The maximum number of components to return in the response. The
# default value is 10.
#
# @option params [String] :next_token
# If the previous call to `ListTrialComponents` didn't return the full
# set of components, the call returns a token for getting the next set
# of components.
#
# @return [Types::ListTrialComponentsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListTrialComponentsResponse#trial_component_summaries #trial_component_summaries} => Array<Types::TrialComponentSummary>
# * {Types::ListTrialComponentsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_trial_components({
# experiment_name: "ExperimentEntityName",
# trial_name: "ExperimentEntityName",
# source_arn: "String256",
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.trial_component_summaries #=> Array
# resp.trial_component_summaries[0].trial_component_name #=> String
# resp.trial_component_summaries[0].trial_component_arn #=> String
# resp.trial_component_summaries[0].display_name #=> String
# resp.trial_component_summaries[0].trial_component_source.source_arn #=> String
# resp.trial_component_summaries[0].trial_component_source.source_type #=> String
# resp.trial_component_summaries[0].status.primary_status #=> String, one of "InProgress", "Completed", "Failed", "Stopping", "Stopped"
# resp.trial_component_summaries[0].status.message #=> String
# resp.trial_component_summaries[0].start_time #=> Time
# resp.trial_component_summaries[0].end_time #=> Time
# resp.trial_component_summaries[0].creation_time #=> Time
# resp.trial_component_summaries[0].created_by.user_profile_arn #=> String
# resp.trial_component_summaries[0].created_by.user_profile_name #=> String
# resp.trial_component_summaries[0].created_by.domain_id #=> String
# resp.trial_component_summaries[0].created_by.iam_identity.arn #=> String
# resp.trial_component_summaries[0].created_by.iam_identity.principal_id #=> String
# resp.trial_component_summaries[0].created_by.iam_identity.source_identity #=> String
# resp.trial_component_summaries[0].last_modified_time #=> Time
# resp.trial_component_summaries[0].last_modified_by.user_profile_arn #=> String
# resp.trial_component_summaries[0].last_modified_by.user_profile_name #=> String
# resp.trial_component_summaries[0].last_modified_by.domain_id #=> String
# resp.trial_component_summaries[0].last_modified_by.iam_identity.arn #=> String
# resp.trial_component_summaries[0].last_modified_by.iam_identity.principal_id #=> String
# resp.trial_component_summaries[0].last_modified_by.iam_identity.source_identity #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListTrialComponents AWS API Documentation
#
# @overload list_trial_components(params = {})
# @param [Hash] params ({})
def list_trial_components(params = {}, options = {})
req = build_request(:list_trial_components, params)
req.send_request(options)
end
# Lists the trials in your account. Specify an experiment name to limit
# the list to the trials that are part of that experiment. Specify a
# trial component name to limit the list to the trials that associated
# with that trial component. The list can be filtered to show only
# trials that were created in a specific time range. The list can be
# sorted by trial name or creation time.
#
# @option params [String] :experiment_name
# A filter that returns only trials that are part of the specified
# experiment.
#
# @option params [String] :trial_component_name
# A filter that returns only trials that are associated with the
# specified trial component.
#
# @option params [Time,DateTime,Date,Integer,String] :created_after
# A filter that returns only trials created after the specified time.
#
# @option params [Time,DateTime,Date,Integer,String] :created_before
# A filter that returns only trials created before the specified time.
#
# @option params [String] :sort_by
# The property used to sort results. The default value is
# `CreationTime`.
#
# @option params [String] :sort_order
# The sort order. The default value is `Descending`.
#
# @option params [Integer] :max_results
# The maximum number of trials to return in the response. The default
# value is 10.
#
# @option params [String] :next_token
# If the previous call to `ListTrials` didn't return the full set of
# trials, the call returns a token for getting the next set of trials.
#
# @return [Types::ListTrialsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListTrialsResponse#trial_summaries #trial_summaries} => Array<Types::TrialSummary>
# * {Types::ListTrialsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_trials({
# experiment_name: "ExperimentEntityName",
# trial_component_name: "ExperimentEntityName",
# created_after: Time.now,
# created_before: Time.now,
# sort_by: "Name", # accepts Name, CreationTime
# sort_order: "Ascending", # accepts Ascending, Descending
# max_results: 1,
# next_token: "NextToken",
# })
#
# @example Response structure
#
# resp.trial_summaries #=> Array
# resp.trial_summaries[0].trial_arn #=> String
# resp.trial_summaries[0].trial_name #=> String
# resp.trial_summaries[0].display_name #=> String
# resp.trial_summaries[0].trial_source.source_arn #=> String
# resp.trial_summaries[0].trial_source.source_type #=> String
# resp.trial_summaries[0].creation_time #=> Time
# resp.trial_summaries[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListTrials AWS API Documentation
#
# @overload list_trials(params = {})
# @param [Hash] params ({})
def list_trials(params = {}, options = {})
req = build_request(:list_trials, params)
req.send_request(options)
end
# Lists user profiles.
#
# @option params [String] :next_token
# If the previous response was truncated, you will receive this token.
# Use it in your next request to receive the next set of results.
#
# @option params [Integer] :max_results
# Returns a list up to a specified limit.
#
# @option params [String] :sort_order
# The sort order for the results. The default is Ascending.
#
# @option params [String] :sort_by
# The parameter by which to sort the results. The default is
# CreationTime.
#
# @option params [String] :domain_id_equals
# A parameter by which to filter the results.
#
# @option params [String] :user_profile_name_contains
# A parameter by which to filter the results.
#
# @return [Types::ListUserProfilesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListUserProfilesResponse#user_profiles #user_profiles} => Array<Types::UserProfileDetails>
# * {Types::ListUserProfilesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_user_profiles({
# next_token: "NextToken",
# max_results: 1,
# sort_order: "Ascending", # accepts Ascending, Descending
# sort_by: "CreationTime", # accepts CreationTime, LastModifiedTime
# domain_id_equals: "DomainId",
# user_profile_name_contains: "UserProfileName",
# })
#
# @example Response structure
#
# resp.user_profiles #=> Array
# resp.user_profiles[0].domain_id #=> String
# resp.user_profiles[0].user_profile_name #=> String
# resp.user_profiles[0].status #=> String, one of "Deleting", "Failed", "InService", "Pending", "Updating", "Update_Failed", "Delete_Failed"
# resp.user_profiles[0].creation_time #=> Time
# resp.user_profiles[0].last_modified_time #=> Time
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListUserProfiles AWS API Documentation
#
# @overload list_user_profiles(params = {})
# @param [Hash] params ({})
def list_user_profiles(params = {}, options = {})
req = build_request(:list_user_profiles, params)
req.send_request(options)
end
# Use this operation to list all private and vendor workforces in an
# Amazon Web Services Region. Note that you can only have one private
# workforce per Amazon Web Services Region.
#
# @option params [String] :sort_by
# Sort workforces using the workforce name or creation date.
#
# @option params [String] :sort_order
# Sort workforces in ascending or descending order.
#
# @option params [String] :name_contains
# A filter you can use to search for workforces using part of the
# workforce name.
#
# @option params [String] :next_token
# A token to resume pagination.
#
# @option params [Integer] :max_results
# The maximum number of workforces returned in the response.
#
# @return [Types::ListWorkforcesResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListWorkforcesResponse#workforces #workforces} => Array<Types::Workforce>
# * {Types::ListWorkforcesResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_workforces({
# sort_by: "Name", # accepts Name, CreateDate
# sort_order: "Ascending", # accepts Ascending, Descending
# name_contains: "WorkforceName",
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.workforces #=> Array
# resp.workforces[0].workforce_name #=> String
# resp.workforces[0].workforce_arn #=> String
# resp.workforces[0].last_updated_date #=> Time
# resp.workforces[0].source_ip_config.cidrs #=> Array
# resp.workforces[0].source_ip_config.cidrs[0] #=> String
# resp.workforces[0].sub_domain #=> String
# resp.workforces[0].cognito_config.user_pool #=> String
# resp.workforces[0].cognito_config.client_id #=> String
# resp.workforces[0].oidc_config.client_id #=> String
# resp.workforces[0].oidc_config.issuer #=> String
# resp.workforces[0].oidc_config.authorization_endpoint #=> String
# resp.workforces[0].oidc_config.token_endpoint #=> String
# resp.workforces[0].oidc_config.user_info_endpoint #=> String
# resp.workforces[0].oidc_config.logout_endpoint #=> String
# resp.workforces[0].oidc_config.jwks_uri #=> String
# resp.workforces[0].create_date #=> Time
# resp.workforces[0].workforce_vpc_config.vpc_id #=> String
# resp.workforces[0].workforce_vpc_config.security_group_ids #=> Array
# resp.workforces[0].workforce_vpc_config.security_group_ids[0] #=> String
# resp.workforces[0].workforce_vpc_config.subnets #=> Array
# resp.workforces[0].workforce_vpc_config.subnets[0] #=> String
# resp.workforces[0].workforce_vpc_config.vpc_endpoint_id #=> String
# resp.workforces[0].status #=> String, one of "Initializing", "Updating", "Deleting", "Failed", "Active"
# resp.workforces[0].failure_reason #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListWorkforces AWS API Documentation
#
# @overload list_workforces(params = {})
# @param [Hash] params ({})
def list_workforces(params = {}, options = {})
req = build_request(:list_workforces, params)
req.send_request(options)
end
# Gets a list of private work teams that you have defined in a region.
# The list may be empty if no work team satisfies the filter specified
# in the `NameContains` parameter.
#
# @option params [String] :sort_by
# The field to sort results by. The default is `CreationTime`.
#
# @option params [String] :sort_order
# The sort order for results. The default is `Ascending`.
#
# @option params [String] :name_contains
# A string in the work team's name. This filter returns only work teams
# whose name contains the specified string.
#
# @option params [String] :next_token
# If the result of the previous `ListWorkteams` request was truncated,
# the response includes a `NextToken`. To retrieve the next set of
# labeling jobs, use the token in the next request.
#
# @option params [Integer] :max_results
# The maximum number of work teams to return in each page of the
# response.
#
# @return [Types::ListWorkteamsResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::ListWorkteamsResponse#workteams #workteams} => Array<Types::Workteam>
# * {Types::ListWorkteamsResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.list_workteams({
# sort_by: "Name", # accepts Name, CreateDate
# sort_order: "Ascending", # accepts Ascending, Descending
# name_contains: "WorkteamName",
# next_token: "NextToken",
# max_results: 1,
# })
#
# @example Response structure
#
# resp.workteams #=> Array
# resp.workteams[0].workteam_name #=> String
# resp.workteams[0].member_definitions #=> Array
# resp.workteams[0].member_definitions[0].cognito_member_definition.user_pool #=> String
# resp.workteams[0].member_definitions[0].cognito_member_definition.user_group #=> String
# resp.workteams[0].member_definitions[0].cognito_member_definition.client_id #=> String
# resp.workteams[0].member_definitions[0].oidc_member_definition.groups #=> Array
# resp.workteams[0].member_definitions[0].oidc_member_definition.groups[0] #=> String
# resp.workteams[0].workteam_arn #=> String
# resp.workteams[0].workforce_arn #=> String
# resp.workteams[0].product_listing_ids #=> Array
# resp.workteams[0].product_listing_ids[0] #=> String
# resp.workteams[0].description #=> String
# resp.workteams[0].sub_domain #=> String
# resp.workteams[0].create_date #=> Time
# resp.workteams[0].last_updated_date #=> Time
# resp.workteams[0].notification_configuration.notification_topic_arn #=> String
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ListWorkteams AWS API Documentation
#
# @overload list_workteams(params = {})
# @param [Hash] params ({})
def list_workteams(params = {}, options = {})
req = build_request(:list_workteams, params)
req.send_request(options)
end
# Adds a resouce policy to control access to a model group. For
# information about resoure policies, see [Identity-based policies and
# resource-based policies][1] in the *Amazon Web Services Identity and
# Access Management User Guide.*.
#
#
#
# [1]: https://docs.aws.amazon.com/IAM/latest/UserGuide/access_policies_identity-vs-resource.html
#
# @option params [required, String] :model_package_group_name
# The name of the model group to add a resource policy to.
#
# @option params [required, String] :resource_policy
# The resource policy for the model group.
#
# @return [Types::PutModelPackageGroupPolicyOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::PutModelPackageGroupPolicyOutput#model_package_group_arn #model_package_group_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.put_model_package_group_policy({
# model_package_group_name: "EntityName", # required
# resource_policy: "PolicyString", # required
# })
#
# @example Response structure
#
# resp.model_package_group_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/PutModelPackageGroupPolicy AWS API Documentation
#
# @overload put_model_package_group_policy(params = {})
# @param [Hash] params ({})
def put_model_package_group_policy(params = {}, options = {})
req = build_request(:put_model_package_group_policy, params)
req.send_request(options)
end
# Use this action to inspect your lineage and discover relationships
# between entities. For more information, see [ Querying Lineage
# Entities][1] in the *Amazon SageMaker Developer Guide*.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/querying-lineage-entities.html
#
# @option params [Array] :start_arns
# A list of resource Amazon Resource Name (ARN) that represent the
# starting point for your lineage query.
#
# @option params [String] :direction
# Associations between lineage entities have a direction. This parameter
# determines the direction from the StartArn(s) that the query
# traverses.
#
# @option params [Boolean] :include_edges
# Setting this value to `True` retrieves not only the entities of
# interest but also the [Associations][1] and lineage entities on the
# path. Set to `False` to only return lineage entities that match your
# query.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/lineage-tracking-entities.html
#
# @option params [Types::QueryFilters] :filters
# A set of filtering parameters that allow you to specify which entities
# should be returned.
#
# * Properties - Key-value pairs to match on the lineage entities'
# properties.
#
# * LineageTypes - A set of lineage entity types to match on. For
# example: `TrialComponent`, `Artifact`, or `Context`.
#
# * CreatedBefore - Filter entities created before this date.
#
# * ModifiedBefore - Filter entities modified before this date.
#
# * ModifiedAfter - Filter entities modified after this date.
#
# @option params [Integer] :max_depth
# The maximum depth in lineage relationships from the `StartArns` that
# are traversed. Depth is a measure of the number of `Associations` from
# the `StartArn` entity to the matched results.
#
# @option params [Integer] :max_results
# Limits the number of vertices in the results. Use the `NextToken` in a
# response to to retrieve the next page of results.
#
# @option params [String] :next_token
# Limits the number of vertices in the request. Use the `NextToken` in a
# response to to retrieve the next page of results.
#
# @return [Types::QueryLineageResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::QueryLineageResponse#vertices #vertices} => Array<Types::Vertex>
# * {Types::QueryLineageResponse#edges #edges} => Array<Types::Edge>
# * {Types::QueryLineageResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.query_lineage({
# start_arns: ["AssociationEntityArn"],
# direction: "Both", # accepts Both, Ascendants, Descendants
# include_edges: false,
# filters: {
# types: ["String40"],
# lineage_types: ["TrialComponent"], # accepts TrialComponent, Artifact, Context, Action
# created_before: Time.now,
# created_after: Time.now,
# modified_before: Time.now,
# modified_after: Time.now,
# properties: {
# "String256" => "String256",
# },
# },
# max_depth: 1,
# max_results: 1,
# next_token: "String8192",
# })
#
# @example Response structure
#
# resp.vertices #=> Array
# resp.vertices[0].arn #=> String
# resp.vertices[0].type #=> String
# resp.vertices[0].lineage_type #=> String, one of "TrialComponent", "Artifact", "Context", "Action"
# resp.edges #=> Array
# resp.edges[0].source_arn #=> String
# resp.edges[0].destination_arn #=> String
# resp.edges[0].association_type #=> String, one of "ContributedTo", "AssociatedWith", "DerivedFrom", "Produced"
# resp.next_token #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/QueryLineage AWS API Documentation
#
# @overload query_lineage(params = {})
# @param [Hash] params ({})
def query_lineage(params = {}, options = {})
req = build_request(:query_lineage, params)
req.send_request(options)
end
# Register devices.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet.
#
# @option params [required, Array] :devices
# A list of devices to register with SageMaker Edge Manager.
#
# @option params [Array] :tags
# The tags associated with devices.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.register_devices({
# device_fleet_name: "EntityName", # required
# devices: [ # required
# {
# device_name: "DeviceName", # required
# description: "DeviceDescription",
# iot_thing_name: "ThingName",
# },
# ],
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/RegisterDevices AWS API Documentation
#
# @overload register_devices(params = {})
# @param [Hash] params ({})
def register_devices(params = {}, options = {})
req = build_request(:register_devices, params)
req.send_request(options)
end
# Renders the UI template so that you can preview the worker's
# experience.
#
# @option params [Types::UiTemplate] :ui_template
# A `Template` object containing the worker UI template to render.
#
# @option params [required, Types::RenderableTask] :task
# A `RenderableTask` object containing a representative task to render.
#
# @option params [required, String] :role_arn
# The Amazon Resource Name (ARN) that has access to the S3 objects that
# are used by the template.
#
# @option params [String] :human_task_ui_arn
# The `HumanTaskUiArn` of the worker UI that you want to render. Do not
# provide a `HumanTaskUiArn` if you use the `UiTemplate` parameter.
#
# See a list of available Human Ui Amazon Resource Names (ARNs) in
# [UiConfig][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_UiConfig.html
#
# @return [Types::RenderUiTemplateResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::RenderUiTemplateResponse#rendered_content #rendered_content} => String
# * {Types::RenderUiTemplateResponse#errors #errors} => Array<Types::RenderingError>
#
# @example Request syntax with placeholder values
#
# resp = client.render_ui_template({
# ui_template: {
# content: "TemplateContent", # required
# },
# task: { # required
# input: "TaskInput", # required
# },
# role_arn: "RoleArn", # required
# human_task_ui_arn: "HumanTaskUiArn",
# })
#
# @example Response structure
#
# resp.rendered_content #=> String
# resp.errors #=> Array
# resp.errors[0].code #=> String
# resp.errors[0].message #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/RenderUiTemplate AWS API Documentation
#
# @overload render_ui_template(params = {})
# @param [Hash] params ({})
def render_ui_template(params = {}, options = {})
req = build_request(:render_ui_template, params)
req.send_request(options)
end
# Retry the execution of the pipeline.
#
# @option params [required, String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @option params [required, String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than once.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @option params [Types::ParallelismConfiguration] :parallelism_configuration
# This configuration, if specified, overrides the parallelism
# configuration of the parent pipeline.
#
# @return [Types::RetryPipelineExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::RetryPipelineExecutionResponse#pipeline_execution_arn #pipeline_execution_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.retry_pipeline_execution({
# pipeline_execution_arn: "PipelineExecutionArn", # required
# client_request_token: "IdempotencyToken", # required
# parallelism_configuration: {
# max_parallel_execution_steps: 1, # required
# },
# })
#
# @example Response structure
#
# resp.pipeline_execution_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/RetryPipelineExecution AWS API Documentation
#
# @overload retry_pipeline_execution(params = {})
# @param [Hash] params ({})
def retry_pipeline_execution(params = {}, options = {})
req = build_request(:retry_pipeline_execution, params)
req.send_request(options)
end
# Finds SageMaker resources that match a search query. Matching
# resources are returned as a list of `SearchRecord` objects in the
# response. You can sort the search results by any resource property in
# a ascending or descending order.
#
# You can query against the following value types: numeric, text,
# Boolean, and timestamp.
#
# The Search API may provide access to otherwise restricted data. See
# [Amazon SageMaker API Permissions: Actions, Permissions, and Resources
# Reference][1] for more information.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/api-permissions-reference.html
#
# @option params [required, String] :resource
# The name of the SageMaker resource to search for.
#
# @option params [Types::SearchExpression] :search_expression
# A Boolean conditional statement. Resources must satisfy this condition
# to be included in search results. You must provide at least one
# subexpression, filter, or nested filter. The maximum number of
# recursive `SubExpressions`, `NestedFilters`, and `Filters` that can be
# included in a `SearchExpression` object is 50.
#
# @option params [String] :sort_by
# The name of the resource property used to sort the `SearchResults`.
# The default is `LastModifiedTime`.
#
# @option params [String] :sort_order
# How `SearchResults` are ordered. Valid values are `Ascending` or
# `Descending`. The default is `Descending`.
#
# @option params [String] :next_token
# If more than `MaxResults` resources match the specified
# `SearchExpression`, the response includes a `NextToken`. The
# `NextToken` can be passed to the next `SearchRequest` to continue
# retrieving results.
#
# @option params [Integer] :max_results
# The maximum number of results to return.
#
# @return [Types::SearchResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::SearchResponse#results #results} => Array<Types::SearchRecord>
# * {Types::SearchResponse#next_token #next_token} => String
#
# The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
#
# @example Request syntax with placeholder values
#
# resp = client.search({
# resource: "TrainingJob", # required, accepts TrainingJob, Experiment, ExperimentTrial, ExperimentTrialComponent, Endpoint, ModelPackage, ModelPackageGroup, Pipeline, PipelineExecution, FeatureGroup, Project, FeatureMetadata, HyperParameterTuningJob, ModelCard, Model
# search_expression: {
# filters: [
# {
# name: "ResourcePropertyName", # required
# operator: "Equals", # accepts Equals, NotEquals, GreaterThan, GreaterThanOrEqualTo, LessThan, LessThanOrEqualTo, Contains, Exists, NotExists, In
# value: "FilterValue",
# },
# ],
# nested_filters: [
# {
# nested_property_name: "ResourcePropertyName", # required
# filters: [ # required
# {
# name: "ResourcePropertyName", # required
# operator: "Equals", # accepts Equals, NotEquals, GreaterThan, GreaterThanOrEqualTo, LessThan, LessThanOrEqualTo, Contains, Exists, NotExists, In
# value: "FilterValue",
# },
# ],
# },
# ],
# sub_expressions: [
# {
# # recursive SearchExpression
# },
# ],
# operator: "And", # accepts And, Or
# },
# sort_by: "ResourcePropertyName",
# sort_order: "Ascending", # accepts Ascending, Descending
# next_token: "NextToken",
# max_results: 1,
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/Search AWS API Documentation
#
# @overload search(params = {})
# @param [Hash] params ({})
def search(params = {}, options = {})
req = build_request(:search, params)
req.send_request(options)
end
# Notifies the pipeline that the execution of a callback step failed,
# along with a message describing why. When a callback step is run, the
# pipeline generates a callback token and includes the token in a
# message sent to Amazon Simple Queue Service (Amazon SQS).
#
# @option params [required, String] :callback_token
# The pipeline generated token from the Amazon SQS queue.
#
# @option params [String] :failure_reason
# A message describing why the step failed.
#
# @option params [String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than one time.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @return [Types::SendPipelineExecutionStepFailureResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::SendPipelineExecutionStepFailureResponse#pipeline_execution_arn #pipeline_execution_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.send_pipeline_execution_step_failure({
# callback_token: "CallbackToken", # required
# failure_reason: "String256",
# client_request_token: "IdempotencyToken",
# })
#
# @example Response structure
#
# resp.pipeline_execution_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/SendPipelineExecutionStepFailure AWS API Documentation
#
# @overload send_pipeline_execution_step_failure(params = {})
# @param [Hash] params ({})
def send_pipeline_execution_step_failure(params = {}, options = {})
req = build_request(:send_pipeline_execution_step_failure, params)
req.send_request(options)
end
# Notifies the pipeline that the execution of a callback step succeeded
# and provides a list of the step's output parameters. When a callback
# step is run, the pipeline generates a callback token and includes the
# token in a message sent to Amazon Simple Queue Service (Amazon SQS).
#
# @option params [required, String] :callback_token
# The pipeline generated token from the Amazon SQS queue.
#
# @option params [Array] :output_parameters
# A list of the output parameters of the callback step.
#
# @option params [String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than one time.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @return [Types::SendPipelineExecutionStepSuccessResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::SendPipelineExecutionStepSuccessResponse#pipeline_execution_arn #pipeline_execution_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.send_pipeline_execution_step_success({
# callback_token: "CallbackToken", # required
# output_parameters: [
# {
# name: "String256", # required
# value: "String1024", # required
# },
# ],
# client_request_token: "IdempotencyToken",
# })
#
# @example Response structure
#
# resp.pipeline_execution_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/SendPipelineExecutionStepSuccess AWS API Documentation
#
# @overload send_pipeline_execution_step_success(params = {})
# @param [Hash] params ({})
def send_pipeline_execution_step_success(params = {}, options = {})
req = build_request(:send_pipeline_execution_step_success, params)
req.send_request(options)
end
# Starts a stage in an edge deployment plan.
#
# @option params [required, String] :edge_deployment_plan_name
# The name of the edge deployment plan to start.
#
# @option params [required, String] :stage_name
# The name of the stage to start.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.start_edge_deployment_stage({
# edge_deployment_plan_name: "EntityName", # required
# stage_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StartEdgeDeploymentStage AWS API Documentation
#
# @overload start_edge_deployment_stage(params = {})
# @param [Hash] params ({})
def start_edge_deployment_stage(params = {}, options = {})
req = build_request(:start_edge_deployment_stage, params)
req.send_request(options)
end
# Starts an inference experiment.
#
# @option params [required, String] :name
# The name of the inference experiment to start.
#
# @return [Types::StartInferenceExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::StartInferenceExperimentResponse#inference_experiment_arn #inference_experiment_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.start_inference_experiment({
# name: "InferenceExperimentName", # required
# })
#
# @example Response structure
#
# resp.inference_experiment_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StartInferenceExperiment AWS API Documentation
#
# @overload start_inference_experiment(params = {})
# @param [Hash] params ({})
def start_inference_experiment(params = {}, options = {})
req = build_request(:start_inference_experiment, params)
req.send_request(options)
end
# Starts a previously stopped monitoring schedule.
#
# By default, when you successfully create a new schedule, the status of
# a monitoring schedule is `scheduled`.
#
#
#
# @option params [required, String] :monitoring_schedule_name
# The name of the schedule to start.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.start_monitoring_schedule({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StartMonitoringSchedule AWS API Documentation
#
# @overload start_monitoring_schedule(params = {})
# @param [Hash] params ({})
def start_monitoring_schedule(params = {}, options = {})
req = build_request(:start_monitoring_schedule, params)
req.send_request(options)
end
# Launches an ML compute instance with the latest version of the
# libraries and attaches your ML storage volume. After configuring the
# notebook instance, SageMaker sets the notebook instance status to
# `InService`. A notebook instance's status must be `InService` before
# you can connect to your Jupyter notebook.
#
# @option params [required, String] :notebook_instance_name
# The name of the notebook instance to start.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.start_notebook_instance({
# notebook_instance_name: "NotebookInstanceName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StartNotebookInstance AWS API Documentation
#
# @overload start_notebook_instance(params = {})
# @param [Hash] params ({})
def start_notebook_instance(params = {}, options = {})
req = build_request(:start_notebook_instance, params)
req.send_request(options)
end
# Starts a pipeline execution.
#
# @option params [required, String] :pipeline_name
# The name of the pipeline.
#
# @option params [String] :pipeline_execution_display_name
# The display name of the pipeline execution.
#
# @option params [Array] :pipeline_parameters
# Contains a list of pipeline parameters. This list can be empty.
#
# @option params [String] :pipeline_execution_description
# The description of the pipeline execution.
#
# @option params [required, String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than once.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @option params [Types::ParallelismConfiguration] :parallelism_configuration
# This configuration, if specified, overrides the parallelism
# configuration of the parent pipeline for this specific run.
#
# @return [Types::StartPipelineExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::StartPipelineExecutionResponse#pipeline_execution_arn #pipeline_execution_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.start_pipeline_execution({
# pipeline_name: "PipelineNameOrArn", # required
# pipeline_execution_display_name: "PipelineExecutionName",
# pipeline_parameters: [
# {
# name: "PipelineParameterName", # required
# value: "String1024", # required
# },
# ],
# pipeline_execution_description: "PipelineExecutionDescription",
# client_request_token: "IdempotencyToken", # required
# parallelism_configuration: {
# max_parallel_execution_steps: 1, # required
# },
# })
#
# @example Response structure
#
# resp.pipeline_execution_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StartPipelineExecution AWS API Documentation
#
# @overload start_pipeline_execution(params = {})
# @param [Hash] params ({})
def start_pipeline_execution(params = {}, options = {})
req = build_request(:start_pipeline_execution, params)
req.send_request(options)
end
# A method for forcing a running job to shut down.
#
# @option params [required, String] :auto_ml_job_name
# The name of the object you are requesting.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_auto_ml_job({
# auto_ml_job_name: "AutoMLJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopAutoMLJob AWS API Documentation
#
# @overload stop_auto_ml_job(params = {})
# @param [Hash] params ({})
def stop_auto_ml_job(params = {}, options = {})
req = build_request(:stop_auto_ml_job, params)
req.send_request(options)
end
# Stops a model compilation job.
#
# To stop a job, Amazon SageMaker sends the algorithm the SIGTERM
# signal. This gracefully shuts the job down. If the job hasn't
# stopped, it sends the SIGKILL signal.
#
# When it receives a `StopCompilationJob` request, Amazon SageMaker
# changes the `CompilationJobStatus` of the job to `Stopping`. After
# Amazon SageMaker stops the job, it sets the `CompilationJobStatus` to
# `Stopped`.
#
# @option params [required, String] :compilation_job_name
# The name of the model compilation job to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_compilation_job({
# compilation_job_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopCompilationJob AWS API Documentation
#
# @overload stop_compilation_job(params = {})
# @param [Hash] params ({})
def stop_compilation_job(params = {}, options = {})
req = build_request(:stop_compilation_job, params)
req.send_request(options)
end
# Stops a stage in an edge deployment plan.
#
# @option params [required, String] :edge_deployment_plan_name
# The name of the edge deployment plan to stop.
#
# @option params [required, String] :stage_name
# The name of the stage to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_edge_deployment_stage({
# edge_deployment_plan_name: "EntityName", # required
# stage_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopEdgeDeploymentStage AWS API Documentation
#
# @overload stop_edge_deployment_stage(params = {})
# @param [Hash] params ({})
def stop_edge_deployment_stage(params = {}, options = {})
req = build_request(:stop_edge_deployment_stage, params)
req.send_request(options)
end
# Request to stop an edge packaging job.
#
# @option params [required, String] :edge_packaging_job_name
# The name of the edge packaging job.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_edge_packaging_job({
# edge_packaging_job_name: "EntityName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopEdgePackagingJob AWS API Documentation
#
# @overload stop_edge_packaging_job(params = {})
# @param [Hash] params ({})
def stop_edge_packaging_job(params = {}, options = {})
req = build_request(:stop_edge_packaging_job, params)
req.send_request(options)
end
# Stops a running hyperparameter tuning job and all running training
# jobs that the tuning job launched.
#
# All model artifacts output from the training jobs are stored in Amazon
# Simple Storage Service (Amazon S3). All data that the training jobs
# write to Amazon CloudWatch Logs are still available in CloudWatch.
# After the tuning job moves to the `Stopped` state, it releases all
# reserved resources for the tuning job.
#
# @option params [required, String] :hyper_parameter_tuning_job_name
# The name of the tuning job to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_hyper_parameter_tuning_job({
# hyper_parameter_tuning_job_name: "HyperParameterTuningJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopHyperParameterTuningJob AWS API Documentation
#
# @overload stop_hyper_parameter_tuning_job(params = {})
# @param [Hash] params ({})
def stop_hyper_parameter_tuning_job(params = {}, options = {})
req = build_request(:stop_hyper_parameter_tuning_job, params)
req.send_request(options)
end
# Stops an inference experiment.
#
# @option params [required, String] :name
# The name of the inference experiment to stop.
#
# @option params [required, Hash] :model_variant_actions
# Array of key-value pairs, with names of variants mapped to actions.
# The possible actions are the following:
#
# * `Promote` - Promote the shadow variant to a production variant
#
# * `Remove` - Delete the variant
#
# * `Retain` - Keep the variant as it is
#
# @option params [Array] :desired_model_variants
# An array of `ModelVariantConfig` objects. There is one for each
# variant that you want to deploy after the inference experiment stops.
# Each `ModelVariantConfig` describes the infrastructure configuration
# for deploying the corresponding variant.
#
# @option params [String] :desired_state
# The desired state of the experiment after stopping. The possible
# states are the following:
#
# * `Completed`: The experiment completed successfully
#
# * `Cancelled`: The experiment was canceled
#
# @option params [String] :reason
# The reason for stopping the experiment.
#
# @return [Types::StopInferenceExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::StopInferenceExperimentResponse#inference_experiment_arn #inference_experiment_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.stop_inference_experiment({
# name: "InferenceExperimentName", # required
# model_variant_actions: { # required
# "ModelVariantName" => "Retain", # accepts Retain, Remove, Promote
# },
# desired_model_variants: [
# {
# model_name: "ModelName", # required
# variant_name: "ModelVariantName", # required
# infrastructure_config: { # required
# infrastructure_type: "RealTimeInference", # required, accepts RealTimeInference
# real_time_inference_config: { # required
# instance_type: "ml.t2.medium", # required, accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge
# instance_count: 1, # required
# },
# },
# },
# ],
# desired_state: "Completed", # accepts Completed, Cancelled
# reason: "InferenceExperimentStatusReason",
# })
#
# @example Response structure
#
# resp.inference_experiment_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopInferenceExperiment AWS API Documentation
#
# @overload stop_inference_experiment(params = {})
# @param [Hash] params ({})
def stop_inference_experiment(params = {}, options = {})
req = build_request(:stop_inference_experiment, params)
req.send_request(options)
end
# Stops an Inference Recommender job.
#
# @option params [required, String] :job_name
# The name of the job you want to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_inference_recommendations_job({
# job_name: "RecommendationJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopInferenceRecommendationsJob AWS API Documentation
#
# @overload stop_inference_recommendations_job(params = {})
# @param [Hash] params ({})
def stop_inference_recommendations_job(params = {}, options = {})
req = build_request(:stop_inference_recommendations_job, params)
req.send_request(options)
end
# Stops a running labeling job. A job that is stopped cannot be
# restarted. Any results obtained before the job is stopped are placed
# in the Amazon S3 output bucket.
#
# @option params [required, String] :labeling_job_name
# The name of the labeling job to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_labeling_job({
# labeling_job_name: "LabelingJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopLabelingJob AWS API Documentation
#
# @overload stop_labeling_job(params = {})
# @param [Hash] params ({})
def stop_labeling_job(params = {}, options = {})
req = build_request(:stop_labeling_job, params)
req.send_request(options)
end
# Stops a previously started monitoring schedule.
#
# @option params [required, String] :monitoring_schedule_name
# The name of the schedule to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_monitoring_schedule({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopMonitoringSchedule AWS API Documentation
#
# @overload stop_monitoring_schedule(params = {})
# @param [Hash] params ({})
def stop_monitoring_schedule(params = {}, options = {})
req = build_request(:stop_monitoring_schedule, params)
req.send_request(options)
end
# Terminates the ML compute instance. Before terminating the instance,
# SageMaker disconnects the ML storage volume from it. SageMaker
# preserves the ML storage volume. SageMaker stops charging you for the
# ML compute instance when you call `StopNotebookInstance`.
#
# To access data on the ML storage volume for a notebook instance that
# has been terminated, call the `StartNotebookInstance` API.
# `StartNotebookInstance` launches another ML compute instance,
# configures it, and attaches the preserved ML storage volume so you can
# continue your work.
#
# @option params [required, String] :notebook_instance_name
# The name of the notebook instance to terminate.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_notebook_instance({
# notebook_instance_name: "NotebookInstanceName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopNotebookInstance AWS API Documentation
#
# @overload stop_notebook_instance(params = {})
# @param [Hash] params ({})
def stop_notebook_instance(params = {}, options = {})
req = build_request(:stop_notebook_instance, params)
req.send_request(options)
end
# Stops a pipeline execution.
#
# **Callback Step**
#
# A pipeline execution won't stop while a callback step is running.
# When you call `StopPipelineExecution` on a pipeline execution with a
# running callback step, SageMaker Pipelines sends an additional Amazon
# SQS message to the specified SQS queue. The body of the SQS message
# contains a "Status" field which is set to "Stopping".
#
# You should add logic to your Amazon SQS message consumer to take any
# needed action (for example, resource cleanup) upon receipt of the
# message followed by a call to `SendPipelineExecutionStepSuccess` or
# `SendPipelineExecutionStepFailure`.
#
# Only when SageMaker Pipelines receives one of these calls will it stop
# the pipeline execution.
#
# **Lambda Step**
#
# A pipeline execution can't be stopped while a lambda step is running
# because the Lambda function invoked by the lambda step can't be
# stopped. If you attempt to stop the execution while the Lambda
# function is running, the pipeline waits for the Lambda function to
# finish or until the timeout is hit, whichever occurs first, and then
# stops. If the Lambda function finishes, the pipeline execution status
# is `Stopped`. If the timeout is hit the pipeline execution status is
# `Failed`.
#
# @option params [required, String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @option params [required, String] :client_request_token
# A unique, case-sensitive identifier that you provide to ensure the
# idempotency of the operation. An idempotent operation completes no
# more than once.
#
# **A suitable default value is auto-generated.** You should normally
# not need to pass this option.**
#
# @return [Types::StopPipelineExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::StopPipelineExecutionResponse#pipeline_execution_arn #pipeline_execution_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.stop_pipeline_execution({
# pipeline_execution_arn: "PipelineExecutionArn", # required
# client_request_token: "IdempotencyToken", # required
# })
#
# @example Response structure
#
# resp.pipeline_execution_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopPipelineExecution AWS API Documentation
#
# @overload stop_pipeline_execution(params = {})
# @param [Hash] params ({})
def stop_pipeline_execution(params = {}, options = {})
req = build_request(:stop_pipeline_execution, params)
req.send_request(options)
end
# Stops a processing job.
#
# @option params [required, String] :processing_job_name
# The name of the processing job to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_processing_job({
# processing_job_name: "ProcessingJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopProcessingJob AWS API Documentation
#
# @overload stop_processing_job(params = {})
# @param [Hash] params ({})
def stop_processing_job(params = {}, options = {})
req = build_request(:stop_processing_job, params)
req.send_request(options)
end
# Stops a training job. To stop a job, SageMaker sends the algorithm the
# `SIGTERM` signal, which delays job termination for 120 seconds.
# Algorithms might use this 120-second window to save the model
# artifacts, so the results of the training is not lost.
#
# When it receives a `StopTrainingJob` request, SageMaker changes the
# status of the job to `Stopping`. After SageMaker stops the job, it
# sets the status to `Stopped`.
#
# @option params [required, String] :training_job_name
# The name of the training job to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_training_job({
# training_job_name: "TrainingJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopTrainingJob AWS API Documentation
#
# @overload stop_training_job(params = {})
# @param [Hash] params ({})
def stop_training_job(params = {}, options = {})
req = build_request(:stop_training_job, params)
req.send_request(options)
end
# Stops a batch transform job.
#
# When Amazon SageMaker receives a `StopTransformJob` request, the
# status of the job changes to `Stopping`. After Amazon SageMaker stops
# the job, the status is set to `Stopped`. When you stop a batch
# transform job before it is completed, Amazon SageMaker doesn't store
# the job's output in Amazon S3.
#
# @option params [required, String] :transform_job_name
# The name of the batch transform job to stop.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.stop_transform_job({
# transform_job_name: "TransformJobName", # required
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/StopTransformJob AWS API Documentation
#
# @overload stop_transform_job(params = {})
# @param [Hash] params ({})
def stop_transform_job(params = {}, options = {})
req = build_request(:stop_transform_job, params)
req.send_request(options)
end
# Updates an action.
#
# @option params [required, String] :action_name
# The name of the action to update.
#
# @option params [String] :description
# The new description for the action.
#
# @option params [String] :status
# The new status for the action.
#
# @option params [Hash] :properties
# The new list of properties. Overwrites the current property list.
#
# @option params [Array] :properties_to_remove
# A list of properties to remove.
#
# @return [Types::UpdateActionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateActionResponse#action_arn #action_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_action({
# action_name: "ExperimentEntityName", # required
# description: "ExperimentDescription",
# status: "Unknown", # accepts Unknown, InProgress, Completed, Failed, Stopping, Stopped
# properties: {
# "StringParameterValue" => "StringParameterValue",
# },
# properties_to_remove: ["StringParameterValue"],
# })
#
# @example Response structure
#
# resp.action_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateAction AWS API Documentation
#
# @overload update_action(params = {})
# @param [Hash] params ({})
def update_action(params = {}, options = {})
req = build_request(:update_action, params)
req.send_request(options)
end
# Updates the properties of an AppImageConfig.
#
# @option params [required, String] :app_image_config_name
# The name of the AppImageConfig to update.
#
# @option params [Types::KernelGatewayImageConfig] :kernel_gateway_image_config
# The new KernelGateway app to run on the image.
#
# @return [Types::UpdateAppImageConfigResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateAppImageConfigResponse#app_image_config_arn #app_image_config_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_app_image_config({
# app_image_config_name: "AppImageConfigName", # required
# kernel_gateway_image_config: {
# kernel_specs: [ # required
# {
# name: "KernelName", # required
# display_name: "KernelDisplayName",
# },
# ],
# file_system_config: {
# mount_path: "MountPath",
# default_uid: 1,
# default_gid: 1,
# },
# },
# })
#
# @example Response structure
#
# resp.app_image_config_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateAppImageConfig AWS API Documentation
#
# @overload update_app_image_config(params = {})
# @param [Hash] params ({})
def update_app_image_config(params = {}, options = {})
req = build_request(:update_app_image_config, params)
req.send_request(options)
end
# Updates an artifact.
#
# @option params [required, String] :artifact_arn
# The Amazon Resource Name (ARN) of the artifact to update.
#
# @option params [String] :artifact_name
# The new name for the artifact.
#
# @option params [Hash] :properties
# The new list of properties. Overwrites the current property list.
#
# @option params [Array] :properties_to_remove
# A list of properties to remove.
#
# @return [Types::UpdateArtifactResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateArtifactResponse#artifact_arn #artifact_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_artifact({
# artifact_arn: "ArtifactArn", # required
# artifact_name: "ExperimentEntityName",
# properties: {
# "StringParameterValue" => "StringParameterValue",
# },
# properties_to_remove: ["StringParameterValue"],
# })
#
# @example Response structure
#
# resp.artifact_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateArtifact AWS API Documentation
#
# @overload update_artifact(params = {})
# @param [Hash] params ({})
def update_artifact(params = {}, options = {})
req = build_request(:update_artifact, params)
req.send_request(options)
end
# Updates the specified Git repository with the specified values.
#
# @option params [required, String] :code_repository_name
# The name of the Git repository to update.
#
# @option params [Types::GitConfigForUpdate] :git_config
# The configuration of the git repository, including the URL and the
# Amazon Resource Name (ARN) of the Amazon Web Services Secrets Manager
# secret that contains the credentials used to access the repository.
# The secret must have a staging label of `AWSCURRENT` and must be in
# the following format:
#
# `\{"username": UserName, "password": Password\}`
#
# @return [Types::UpdateCodeRepositoryOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateCodeRepositoryOutput#code_repository_arn #code_repository_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_code_repository({
# code_repository_name: "EntityName", # required
# git_config: {
# secret_arn: "SecretArn",
# },
# })
#
# @example Response structure
#
# resp.code_repository_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateCodeRepository AWS API Documentation
#
# @overload update_code_repository(params = {})
# @param [Hash] params ({})
def update_code_repository(params = {}, options = {})
req = build_request(:update_code_repository, params)
req.send_request(options)
end
# Updates a context.
#
# @option params [required, String] :context_name
# The name of the context to update.
#
# @option params [String] :description
# The new description for the context.
#
# @option params [Hash] :properties
# The new list of properties. Overwrites the current property list.
#
# @option params [Array] :properties_to_remove
# A list of properties to remove.
#
# @return [Types::UpdateContextResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateContextResponse#context_arn #context_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_context({
# context_name: "ExperimentEntityName", # required
# description: "ExperimentDescription",
# properties: {
# "StringParameterValue" => "StringParameterValue",
# },
# properties_to_remove: ["StringParameterValue"],
# })
#
# @example Response structure
#
# resp.context_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateContext AWS API Documentation
#
# @overload update_context(params = {})
# @param [Hash] params ({})
def update_context(params = {}, options = {})
req = build_request(:update_context, params)
req.send_request(options)
end
# Updates a fleet of devices.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet.
#
# @option params [String] :role_arn
# The Amazon Resource Name (ARN) of the device.
#
# @option params [String] :description
# Description of the fleet.
#
# @option params [required, Types::EdgeOutputConfig] :output_config
# Output configuration for storing sample data collected by the fleet.
#
# @option params [Boolean] :enable_iot_role_alias
# Whether to create an Amazon Web Services IoT Role Alias during device
# fleet creation. The name of the role alias generated will match this
# pattern: "SageMakerEdge-\\\{DeviceFleetName\\}".
#
# For example, if your device fleet is called "demo-fleet", the name
# of the role alias will be "SageMakerEdge-demo-fleet".
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.update_device_fleet({
# device_fleet_name: "EntityName", # required
# role_arn: "RoleArn",
# description: "DeviceFleetDescription",
# output_config: { # required
# s3_output_location: "S3Uri", # required
# kms_key_id: "KmsKeyId",
# preset_deployment_type: "GreengrassV2Component", # accepts GreengrassV2Component
# preset_deployment_config: "String",
# },
# enable_iot_role_alias: false,
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateDeviceFleet AWS API Documentation
#
# @overload update_device_fleet(params = {})
# @param [Hash] params ({})
def update_device_fleet(params = {}, options = {})
req = build_request(:update_device_fleet, params)
req.send_request(options)
end
# Updates one or more devices in a fleet.
#
# @option params [required, String] :device_fleet_name
# The name of the fleet the devices belong to.
#
# @option params [required, Array] :devices
# List of devices to register with Edge Manager agent.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.update_devices({
# device_fleet_name: "EntityName", # required
# devices: [ # required
# {
# device_name: "DeviceName", # required
# description: "DeviceDescription",
# iot_thing_name: "ThingName",
# },
# ],
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateDevices AWS API Documentation
#
# @overload update_devices(params = {})
# @param [Hash] params ({})
def update_devices(params = {}, options = {})
req = build_request(:update_devices, params)
req.send_request(options)
end
# Updates the default settings for new user profiles in the domain.
#
# @option params [required, String] :domain_id
# The ID of the domain to be updated.
#
# @option params [Types::UserSettings] :default_user_settings
# A collection of settings.
#
# @option params [Types::DomainSettingsForUpdate] :domain_settings_for_update
# A collection of `DomainSettings` configuration values to update.
#
# @option params [Types::DefaultSpaceSettings] :default_space_settings
# The default settings used to create a space within the Domain.
#
# @option params [String] :app_security_group_management
# The entity that creates and manages the required security groups for
# inter-app communication in `VPCOnly` mode. Required when
# `CreateDomain.AppNetworkAccessType` is `VPCOnly` and
# `DomainSettings.RStudioServerProDomainSettings.DomainExecutionRoleArn`
# is provided.
#
# @return [Types::UpdateDomainResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateDomainResponse#domain_arn #domain_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_domain({
# domain_id: "DomainId", # required
# default_user_settings: {
# execution_role: "RoleArn",
# security_groups: ["SecurityGroupId"],
# sharing_settings: {
# notebook_output_option: "Allowed", # accepts Allowed, Disabled
# s3_output_path: "S3Uri",
# s3_kms_key_id: "KmsKeyId",
# },
# jupyter_server_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# code_repositories: [
# {
# repository_url: "RepositoryUrl", # required
# },
# ],
# },
# kernel_gateway_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# },
# tensor_board_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# },
# r_studio_server_pro_app_settings: {
# access_status: "ENABLED", # accepts ENABLED, DISABLED
# user_group: "R_STUDIO_ADMIN", # accepts R_STUDIO_ADMIN, R_STUDIO_USER
# },
# r_session_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# },
# canvas_app_settings: {
# time_series_forecasting_settings: {
# status: "ENABLED", # accepts ENABLED, DISABLED
# amazon_forecast_role_arn: "RoleArn",
# },
# model_register_settings: {
# status: "ENABLED", # accepts ENABLED, DISABLED
# cross_account_model_register_role_arn: "RoleArn",
# },
# },
# },
# domain_settings_for_update: {
# r_studio_server_pro_domain_settings_for_update: {
# domain_execution_role_arn: "RoleArn", # required
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# r_studio_connect_url: "String",
# r_studio_package_manager_url: "String",
# },
# execution_role_identity_config: "USER_PROFILE_NAME", # accepts USER_PROFILE_NAME, DISABLED
# security_group_ids: ["SecurityGroupId"],
# },
# default_space_settings: {
# execution_role: "RoleArn",
# security_groups: ["SecurityGroupId"],
# jupyter_server_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# code_repositories: [
# {
# repository_url: "RepositoryUrl", # required
# },
# ],
# },
# kernel_gateway_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# },
# },
# app_security_group_management: "Service", # accepts Service, Customer
# })
#
# @example Response structure
#
# resp.domain_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateDomain AWS API Documentation
#
# @overload update_domain(params = {})
# @param [Hash] params ({})
def update_domain(params = {}, options = {})
req = build_request(:update_domain, params)
req.send_request(options)
end
# Deploys the new `EndpointConfig` specified in the request, switches to
# using newly created endpoint, and then deletes resources provisioned
# for the endpoint using the previous `EndpointConfig` (there is no
# availability loss).
#
# When SageMaker receives the request, it sets the endpoint status to
# `Updating`. After updating the endpoint, it sets the status to
# `InService`. To check the status of an endpoint, use the
# [DescribeEndpoint][1] API.
#
# You must not delete an `EndpointConfig` in use by an endpoint that is
# live or while the `UpdateEndpoint` or `CreateEndpoint` operations are
# being performed on the endpoint. To update an endpoint, you must
# create a new `EndpointConfig`.
#
# If you delete the `EndpointConfig` of an endpoint that is active or
# being created or updated you may lose visibility into the instance
# type the endpoint is using. The endpoint must be deleted in order to
# stop incurring charges.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeEndpoint.html
#
# @option params [required, String] :endpoint_name
# The name of the endpoint whose configuration you want to update.
#
# @option params [required, String] :endpoint_config_name
# The name of the new endpoint configuration.
#
# @option params [Boolean] :retain_all_variant_properties
# When updating endpoint resources, enables or disables the retention of
# [variant properties][1], such as the instance count or the variant
# weight. To retain the variant properties of an endpoint when updating
# it, set `RetainAllVariantProperties` to `true`. To use the variant
# properties specified in a new `EndpointConfig` call when updating an
# endpoint, set `RetainAllVariantProperties` to `false`. The default is
# `false`.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_VariantProperty.html
#
# @option params [Array] :exclude_retained_variant_properties
# When you are updating endpoint resources with
# `RetainAllVariantProperties`, whose value is set to `true`,
# `ExcludeRetainedVariantProperties` specifies the list of type
# [VariantProperty][1] to override with the values provided by
# `EndpointConfig`. If you don't specify a value for
# `ExcludeRetainedVariantProperties`, no variant properties are
# overridden.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_VariantProperty.html
#
# @option params [Types::DeploymentConfig] :deployment_config
# The deployment configuration for an endpoint, which contains the
# desired deployment strategy and rollback configurations.
#
# @option params [Boolean] :retain_deployment_config
# Specifies whether to reuse the last deployment configuration. The
# default value is false (the configuration is not reused).
#
# @return [Types::UpdateEndpointOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateEndpointOutput#endpoint_arn #endpoint_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_endpoint({
# endpoint_name: "EndpointName", # required
# endpoint_config_name: "EndpointConfigName", # required
# retain_all_variant_properties: false,
# exclude_retained_variant_properties: [
# {
# variant_property_type: "DesiredInstanceCount", # required, accepts DesiredInstanceCount, DesiredWeight, DataCaptureConfig
# },
# ],
# deployment_config: {
# blue_green_update_policy: { # required
# traffic_routing_configuration: { # required
# type: "ALL_AT_ONCE", # required, accepts ALL_AT_ONCE, CANARY, LINEAR
# wait_interval_in_seconds: 1, # required
# canary_size: {
# type: "INSTANCE_COUNT", # required, accepts INSTANCE_COUNT, CAPACITY_PERCENT
# value: 1, # required
# },
# linear_step_size: {
# type: "INSTANCE_COUNT", # required, accepts INSTANCE_COUNT, CAPACITY_PERCENT
# value: 1, # required
# },
# },
# termination_wait_in_seconds: 1,
# maximum_execution_timeout_in_seconds: 1,
# },
# auto_rollback_configuration: {
# alarms: [
# {
# alarm_name: "AlarmName",
# },
# ],
# },
# },
# retain_deployment_config: false,
# })
#
# @example Response structure
#
# resp.endpoint_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateEndpoint AWS API Documentation
#
# @overload update_endpoint(params = {})
# @param [Hash] params ({})
def update_endpoint(params = {}, options = {})
req = build_request(:update_endpoint, params)
req.send_request(options)
end
# Updates variant weight of one or more variants associated with an
# existing endpoint, or capacity of one variant associated with an
# existing endpoint. When it receives the request, SageMaker sets the
# endpoint status to `Updating`. After updating the endpoint, it sets
# the status to `InService`. To check the status of an endpoint, use the
# [DescribeEndpoint][1] API.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeEndpoint.html
#
# @option params [required, String] :endpoint_name
# The name of an existing SageMaker endpoint.
#
# @option params [required, Array] :desired_weights_and_capacities
# An object that provides new capacity and weight values for a variant.
#
# @return [Types::UpdateEndpointWeightsAndCapacitiesOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateEndpointWeightsAndCapacitiesOutput#endpoint_arn #endpoint_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_endpoint_weights_and_capacities({
# endpoint_name: "EndpointName", # required
# desired_weights_and_capacities: [ # required
# {
# variant_name: "VariantName", # required
# desired_weight: 1.0,
# desired_instance_count: 1,
# },
# ],
# })
#
# @example Response structure
#
# resp.endpoint_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateEndpointWeightsAndCapacities AWS API Documentation
#
# @overload update_endpoint_weights_and_capacities(params = {})
# @param [Hash] params ({})
def update_endpoint_weights_and_capacities(params = {}, options = {})
req = build_request(:update_endpoint_weights_and_capacities, params)
req.send_request(options)
end
# Adds, updates, or removes the description of an experiment. Updates
# the display name of an experiment.
#
# @option params [required, String] :experiment_name
# The name of the experiment to update.
#
# @option params [String] :display_name
# The name of the experiment as displayed. The name doesn't need to be
# unique. If `DisplayName` isn't specified, `ExperimentName` is
# displayed.
#
# @option params [String] :description
# The description of the experiment.
#
# @return [Types::UpdateExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateExperimentResponse#experiment_arn #experiment_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_experiment({
# experiment_name: "ExperimentEntityName", # required
# display_name: "ExperimentEntityName",
# description: "ExperimentDescription",
# })
#
# @example Response structure
#
# resp.experiment_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateExperiment AWS API Documentation
#
# @overload update_experiment(params = {})
# @param [Hash] params ({})
def update_experiment(params = {}, options = {})
req = build_request(:update_experiment, params)
req.send_request(options)
end
# Updates the feature group.
#
# @option params [required, String] :feature_group_name
# The name of the feature group that you're updating.
#
# @option params [Array] :feature_additions
# Updates the feature group. Updating a feature group is an asynchronous
# operation. When you get an HTTP 200 response, you've made a valid
# request. It takes some time after you've made a valid request for
# Feature Store to update the feature group.
#
# @return [Types::UpdateFeatureGroupResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateFeatureGroupResponse#feature_group_arn #feature_group_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_feature_group({
# feature_group_name: "FeatureGroupName", # required
# feature_additions: [
# {
# feature_name: "FeatureName",
# feature_type: "Integral", # accepts Integral, Fractional, String
# },
# ],
# })
#
# @example Response structure
#
# resp.feature_group_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateFeatureGroup AWS API Documentation
#
# @overload update_feature_group(params = {})
# @param [Hash] params ({})
def update_feature_group(params = {}, options = {})
req = build_request(:update_feature_group, params)
req.send_request(options)
end
# Updates the description and parameters of the feature group.
#
# @option params [required, String] :feature_group_name
# The name of the feature group containing the feature that you're
# updating.
#
# @option params [required, String] :feature_name
# The name of the feature that you're updating.
#
# @option params [String] :description
# A description that you can write to better describe the feature.
#
# @option params [Array] :parameter_additions
# A list of key-value pairs that you can add to better describe the
# feature.
#
# @option params [Array] :parameter_removals
# A list of parameter keys that you can specify to remove parameters
# that describe your feature.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.update_feature_metadata({
# feature_group_name: "FeatureGroupName", # required
# feature_name: "FeatureName", # required
# description: "FeatureDescription",
# parameter_additions: [
# {
# key: "FeatureParameterKey",
# value: "FeatureParameterValue",
# },
# ],
# parameter_removals: ["FeatureParameterKey"],
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateFeatureMetadata AWS API Documentation
#
# @overload update_feature_metadata(params = {})
# @param [Hash] params ({})
def update_feature_metadata(params = {}, options = {})
req = build_request(:update_feature_metadata, params)
req.send_request(options)
end
# Update a hub.
#
# Hub APIs are only callable through SageMaker Studio.
#
#
#
# @option params [required, String] :hub_name
# The name of the hub to update.
#
# @option params [String] :hub_description
# A description of the updated hub.
#
# @option params [String] :hub_display_name
# The display name of the hub.
#
# @option params [Array] :hub_search_keywords
# The searchable keywords for the hub.
#
# @return [Types::UpdateHubResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateHubResponse#hub_arn #hub_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_hub({
# hub_name: "HubName", # required
# hub_description: "HubDescription",
# hub_display_name: "HubDisplayName",
# hub_search_keywords: ["HubSearchKeyword"],
# })
#
# @example Response structure
#
# resp.hub_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateHub AWS API Documentation
#
# @overload update_hub(params = {})
# @param [Hash] params ({})
def update_hub(params = {}, options = {})
req = build_request(:update_hub, params)
req.send_request(options)
end
# Updates the properties of a SageMaker image. To change the image's
# tags, use the [AddTags][1] and [DeleteTags][2] APIs.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_AddTags.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DeleteTags.html
#
# @option params [Array] :delete_properties
# A list of properties to delete. Only the `Description` and
# `DisplayName` properties can be deleted.
#
# @option params [String] :description
# The new description for the image.
#
# @option params [String] :display_name
# The new display name for the image.
#
# @option params [required, String] :image_name
# The name of the image to update.
#
# @option params [String] :role_arn
# The new ARN for the IAM role that enables Amazon SageMaker to perform
# tasks on your behalf.
#
# @return [Types::UpdateImageResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateImageResponse#image_arn #image_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_image({
# delete_properties: ["ImageDeleteProperty"],
# description: "ImageDescription",
# display_name: "ImageDisplayName",
# image_name: "ImageName", # required
# role_arn: "RoleArn",
# })
#
# @example Response structure
#
# resp.image_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateImage AWS API Documentation
#
# @overload update_image(params = {})
# @param [Hash] params ({})
def update_image(params = {}, options = {})
req = build_request(:update_image, params)
req.send_request(options)
end
# Updates the properties of a SageMaker image version.
#
# @option params [required, String] :image_name
# The name of the image.
#
# @option params [String] :alias
# The alias of the image version.
#
# @option params [Integer] :version
# The version of the image.
#
# @option params [Array] :aliases_to_add
# A list of aliases to add.
#
# @option params [Array] :aliases_to_delete
# A list of aliases to delete.
#
# @option params [String] :vendor_guidance
# The availability of the image version specified by the maintainer.
#
# * `NOT_PROVIDED`: The maintainers did not provide a status for image
# version stability.
#
# * `STABLE`: The image version is stable.
#
# * `TO_BE_ARCHIVED`: The image version is set to be archived. Custom
# image versions that are set to be archived are automatically
# archived after three months.
#
# * `ARCHIVED`: The image version is archived. Archived image versions
# are not searchable and are no longer actively supported.
#
# @option params [String] :job_type
# Indicates SageMaker job type compatibility.
#
# * `TRAINING`: The image version is compatible with SageMaker training
# jobs.
#
# * `INFERENCE`: The image version is compatible with SageMaker
# inference jobs.
#
# * `NOTEBOOK_KERNEL`: The image version is compatible with SageMaker
# notebook kernels.
#
# @option params [String] :ml_framework
# The machine learning framework vended in the image version.
#
# @option params [String] :programming_lang
# The supported programming language and its version.
#
# @option params [String] :processor
# Indicates CPU or GPU compatibility.
#
# * `CPU`: The image version is compatible with CPU.
#
# * `GPU`: The image version is compatible with GPU.
#
# @option params [Boolean] :horovod
# Indicates Horovod compatibility.
#
# @option params [String] :release_notes
# The maintainer description of the image version.
#
# @return [Types::UpdateImageVersionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateImageVersionResponse#image_version_arn #image_version_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_image_version({
# image_name: "ImageName", # required
# alias: "SageMakerImageVersionAlias",
# version: 1,
# aliases_to_add: ["SageMakerImageVersionAlias"],
# aliases_to_delete: ["SageMakerImageVersionAlias"],
# vendor_guidance: "NOT_PROVIDED", # accepts NOT_PROVIDED, STABLE, TO_BE_ARCHIVED, ARCHIVED
# job_type: "TRAINING", # accepts TRAINING, INFERENCE, NOTEBOOK_KERNEL
# ml_framework: "MLFramework",
# programming_lang: "ProgrammingLang",
# processor: "CPU", # accepts CPU, GPU
# horovod: false,
# release_notes: "ReleaseNotes",
# })
#
# @example Response structure
#
# resp.image_version_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateImageVersion AWS API Documentation
#
# @overload update_image_version(params = {})
# @param [Hash] params ({})
def update_image_version(params = {}, options = {})
req = build_request(:update_image_version, params)
req.send_request(options)
end
# Updates an inference experiment that you created. The status of the
# inference experiment has to be either `Created`, `Running`. For more
# information on the status of an inference experiment, see
# [DescribeInferenceExperiment][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeInferenceExperiment.html
#
# @option params [required, String] :name
# The name of the inference experiment to be updated.
#
# @option params [Types::InferenceExperimentSchedule] :schedule
# The duration for which the inference experiment will run. If the
# status of the inference experiment is `Created`, then you can update
# both the start and end dates. If the status of the inference
# experiment is `Running`, then you can update only the end date.
#
# @option params [String] :description
# The description of the inference experiment.
#
# @option params [Array] :model_variants
# An array of `ModelVariantConfig` objects. There is one for each
# variant, whose infrastructure configuration you want to update.
#
# @option params [Types::InferenceExperimentDataStorageConfig] :data_storage_config
# The Amazon S3 location and configuration for storing inference request
# and response data.
#
# @option params [Types::ShadowModeConfig] :shadow_mode_config
# The configuration of `ShadowMode` inference experiment type. Use this
# field to specify a production variant which takes all the inference
# requests, and a shadow variant to which Amazon SageMaker replicates a
# percentage of the inference requests. For the shadow variant also
# specify the percentage of requests that Amazon SageMaker replicates.
#
# @return [Types::UpdateInferenceExperimentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateInferenceExperimentResponse#inference_experiment_arn #inference_experiment_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_inference_experiment({
# name: "InferenceExperimentName", # required
# schedule: {
# start_time: Time.now,
# end_time: Time.now,
# },
# description: "InferenceExperimentDescription",
# model_variants: [
# {
# model_name: "ModelName", # required
# variant_name: "ModelVariantName", # required
# infrastructure_config: { # required
# infrastructure_type: "RealTimeInference", # required, accepts RealTimeInference
# real_time_inference_config: { # required
# instance_type: "ml.t2.medium", # required, accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge
# instance_count: 1, # required
# },
# },
# },
# ],
# data_storage_config: {
# destination: "DestinationS3Uri", # required
# kms_key: "KmsKeyId",
# content_type: {
# csv_content_types: ["CsvContentType"],
# json_content_types: ["JsonContentType"],
# },
# },
# shadow_mode_config: {
# source_model_variant_name: "ModelVariantName", # required
# shadow_model_variants: [ # required
# {
# shadow_model_variant_name: "ModelVariantName", # required
# sampling_percentage: 1, # required
# },
# ],
# },
# })
#
# @example Response structure
#
# resp.inference_experiment_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateInferenceExperiment AWS API Documentation
#
# @overload update_inference_experiment(params = {})
# @param [Hash] params ({})
def update_inference_experiment(params = {}, options = {})
req = build_request(:update_inference_experiment, params)
req.send_request(options)
end
# Update an Amazon SageMaker Model Card.
#
# You cannot update both model card content and model card status in a
# single call.
#
# @option params [required, String] :model_card_name
# The name of the model card to update.
#
# @option params [String] :content
# The updated model card content. Content must be in [model card JSON
# schema][1] and provided as a string.
#
# When updating model card content, be sure to include the full content
# and not just updated content.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/model-cards.html#model-cards-json-schema
#
# @option params [String] :model_card_status
# The approval status of the model card within your organization.
# Different organizations might have different criteria for model card
# review and approval.
#
# * `Draft`: The model card is a work in progress.
#
# * `PendingReview`: The model card is pending review.
#
# * `Approved`: The model card is approved.
#
# * `Archived`: The model card is archived. No more updates should be
# made to the model card, but it can still be exported.
#
# @return [Types::UpdateModelCardResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateModelCardResponse#model_card_arn #model_card_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_model_card({
# model_card_name: "EntityName", # required
# content: "ModelCardContent",
# model_card_status: "Draft", # accepts Draft, PendingReview, Approved, Archived
# })
#
# @example Response structure
#
# resp.model_card_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateModelCard AWS API Documentation
#
# @overload update_model_card(params = {})
# @param [Hash] params ({})
def update_model_card(params = {}, options = {})
req = build_request(:update_model_card, params)
req.send_request(options)
end
# Updates a versioned model.
#
# @option params [required, String] :model_package_arn
# The Amazon Resource Name (ARN) of the model package.
#
# @option params [String] :model_approval_status
# The approval status of the model.
#
# @option params [String] :approval_description
# A description for the approval status of the model.
#
# @option params [Hash] :customer_metadata_properties
# The metadata properties associated with the model package versions.
#
# @option params [Array] :customer_metadata_properties_to_remove
# The metadata properties associated with the model package versions to
# remove.
#
# @option params [Array] :additional_inference_specifications_to_add
# An array of additional Inference Specification objects to be added to
# the existing array additional Inference Specification. Total number of
# additional Inference Specifications can not exceed 15. Each additional
# Inference Specification specifies artifacts based on this model
# package that can be used on inference endpoints. Generally used with
# SageMaker Neo to store the compiled artifacts.
#
# @return [Types::UpdateModelPackageOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateModelPackageOutput#model_package_arn #model_package_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_model_package({
# model_package_arn: "ModelPackageArn", # required
# model_approval_status: "Approved", # accepts Approved, Rejected, PendingManualApproval
# approval_description: "ApprovalDescription",
# customer_metadata_properties: {
# "CustomerMetadataKey" => "CustomerMetadataValue",
# },
# customer_metadata_properties_to_remove: ["CustomerMetadataKey"],
# additional_inference_specifications_to_add: [
# {
# name: "EntityName", # required
# description: "EntityDescription",
# containers: [ # required
# {
# container_hostname: "ContainerHostname",
# image: "ContainerImage", # required
# image_digest: "ImageDigest",
# model_data_url: "Url",
# product_id: "ProductId",
# environment: {
# "EnvironmentKey" => "EnvironmentValue",
# },
# model_input: {
# data_input_config: "DataInputConfig", # required
# },
# framework: "String",
# framework_version: "ModelPackageFrameworkVersion",
# nearest_model_name: "String",
# },
# ],
# supported_transform_instance_types: ["ml.m4.xlarge"], # accepts ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# supported_realtime_inference_instance_types: ["ml.t2.medium"], # accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.12xlarge, ml.m5d.24xlarge, ml.c4.large, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.large, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.12xlarge, ml.r5.24xlarge, ml.r5d.large, ml.r5d.xlarge, ml.r5d.2xlarge, ml.r5d.4xlarge, ml.r5d.12xlarge, ml.r5d.24xlarge, ml.inf1.xlarge, ml.inf1.2xlarge, ml.inf1.6xlarge, ml.inf1.24xlarge, ml.c6i.large, ml.c6i.xlarge, ml.c6i.2xlarge, ml.c6i.4xlarge, ml.c6i.8xlarge, ml.c6i.12xlarge, ml.c6i.16xlarge, ml.c6i.24xlarge, ml.c6i.32xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.12xlarge, ml.g5.16xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.p4d.24xlarge, ml.c7g.large, ml.c7g.xlarge, ml.c7g.2xlarge, ml.c7g.4xlarge, ml.c7g.8xlarge, ml.c7g.12xlarge, ml.c7g.16xlarge, ml.m6g.large, ml.m6g.xlarge, ml.m6g.2xlarge, ml.m6g.4xlarge, ml.m6g.8xlarge, ml.m6g.12xlarge, ml.m6g.16xlarge, ml.m6gd.large, ml.m6gd.xlarge, ml.m6gd.2xlarge, ml.m6gd.4xlarge, ml.m6gd.8xlarge, ml.m6gd.12xlarge, ml.m6gd.16xlarge, ml.c6g.large, ml.c6g.xlarge, ml.c6g.2xlarge, ml.c6g.4xlarge, ml.c6g.8xlarge, ml.c6g.12xlarge, ml.c6g.16xlarge, ml.c6gd.large, ml.c6gd.xlarge, ml.c6gd.2xlarge, ml.c6gd.4xlarge, ml.c6gd.8xlarge, ml.c6gd.12xlarge, ml.c6gd.16xlarge, ml.c6gn.large, ml.c6gn.xlarge, ml.c6gn.2xlarge, ml.c6gn.4xlarge, ml.c6gn.8xlarge, ml.c6gn.12xlarge, ml.c6gn.16xlarge, ml.r6g.large, ml.r6g.xlarge, ml.r6g.2xlarge, ml.r6g.4xlarge, ml.r6g.8xlarge, ml.r6g.12xlarge, ml.r6g.16xlarge, ml.r6gd.large, ml.r6gd.xlarge, ml.r6gd.2xlarge, ml.r6gd.4xlarge, ml.r6gd.8xlarge, ml.r6gd.12xlarge, ml.r6gd.16xlarge, ml.p4de.24xlarge, ml.trn1.2xlarge, ml.trn1.32xlarge, ml.inf2.xlarge, ml.inf2.8xlarge, ml.inf2.24xlarge, ml.inf2.48xlarge
# supported_content_types: ["ContentType"],
# supported_response_mime_types: ["ResponseMIMEType"],
# },
# ],
# })
#
# @example Response structure
#
# resp.model_package_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateModelPackage AWS API Documentation
#
# @overload update_model_package(params = {})
# @param [Hash] params ({})
def update_model_package(params = {}, options = {})
req = build_request(:update_model_package, params)
req.send_request(options)
end
# Update the parameters of a model monitor alert.
#
# @option params [required, String] :monitoring_schedule_name
# The name of a monitoring schedule.
#
# @option params [required, String] :monitoring_alert_name
# The name of a monitoring alert.
#
# @option params [required, Integer] :datapoints_to_alert
# Within `EvaluationPeriod`, how many execution failures will raise an
# alert.
#
# @option params [required, Integer] :evaluation_period
# The number of most recent monitoring executions to consider when
# evaluating alert status.
#
# @return [Types::UpdateMonitoringAlertResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateMonitoringAlertResponse#monitoring_schedule_arn #monitoring_schedule_arn} => String
# * {Types::UpdateMonitoringAlertResponse#monitoring_alert_name #monitoring_alert_name} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_monitoring_alert({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# monitoring_alert_name: "MonitoringAlertName", # required
# datapoints_to_alert: 1, # required
# evaluation_period: 1, # required
# })
#
# @example Response structure
#
# resp.monitoring_schedule_arn #=> String
# resp.monitoring_alert_name #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateMonitoringAlert AWS API Documentation
#
# @overload update_monitoring_alert(params = {})
# @param [Hash] params ({})
def update_monitoring_alert(params = {}, options = {})
req = build_request(:update_monitoring_alert, params)
req.send_request(options)
end
# Updates a previously created schedule.
#
# @option params [required, String] :monitoring_schedule_name
# The name of the monitoring schedule. The name must be unique within an
# Amazon Web Services Region within an Amazon Web Services account.
#
# @option params [required, Types::MonitoringScheduleConfig] :monitoring_schedule_config
# The configuration object that specifies the monitoring schedule and
# defines the monitoring job.
#
# @return [Types::UpdateMonitoringScheduleResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateMonitoringScheduleResponse#monitoring_schedule_arn #monitoring_schedule_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_monitoring_schedule({
# monitoring_schedule_name: "MonitoringScheduleName", # required
# monitoring_schedule_config: { # required
# schedule_config: {
# schedule_expression: "ScheduleExpression", # required
# },
# monitoring_job_definition: {
# baseline_config: {
# baselining_job_name: "ProcessingJobName",
# constraints_resource: {
# s3_uri: "S3Uri",
# },
# statistics_resource: {
# s3_uri: "S3Uri",
# },
# },
# monitoring_inputs: [ # required
# {
# endpoint_input: {
# endpoint_name: "EndpointName", # required
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# batch_transform_input: {
# data_captured_destination_s3_uri: "DestinationS3Uri", # required
# dataset_format: { # required
# csv: {
# header: false,
# },
# json: {
# line: false,
# },
# parquet: {
# },
# },
# local_path: "ProcessingLocalPath", # required
# s3_input_mode: "Pipe", # accepts Pipe, File
# s3_data_distribution_type: "FullyReplicated", # accepts FullyReplicated, ShardedByS3Key
# features_attribute: "String",
# inference_attribute: "String",
# probability_attribute: "String",
# probability_threshold_attribute: 1.0,
# start_time_offset: "MonitoringTimeOffsetString",
# end_time_offset: "MonitoringTimeOffsetString",
# },
# },
# ],
# monitoring_output_config: { # required
# monitoring_outputs: [ # required
# {
# s3_output: { # required
# s3_uri: "MonitoringS3Uri", # required
# local_path: "ProcessingLocalPath", # required
# s3_upload_mode: "Continuous", # accepts Continuous, EndOfJob
# },
# },
# ],
# kms_key_id: "KmsKeyId",
# },
# monitoring_resources: { # required
# cluster_config: { # required
# instance_count: 1, # required
# instance_type: "ml.t3.medium", # required, accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1, # required
# volume_kms_key_id: "KmsKeyId",
# },
# },
# monitoring_app_specification: { # required
# image_uri: "ImageUri", # required
# container_entrypoint: ["ContainerEntrypointString"],
# container_arguments: ["ContainerArgument"],
# record_preprocessor_source_uri: "S3Uri",
# post_analytics_processor_source_uri: "S3Uri",
# },
# stopping_condition: {
# max_runtime_in_seconds: 1, # required
# },
# environment: {
# "ProcessingEnvironmentKey" => "ProcessingEnvironmentValue",
# },
# network_config: {
# enable_inter_container_traffic_encryption: false,
# enable_network_isolation: false,
# vpc_config: {
# security_group_ids: ["SecurityGroupId"], # required
# subnets: ["SubnetId"], # required
# },
# },
# role_arn: "RoleArn", # required
# },
# monitoring_job_definition_name: "MonitoringJobDefinitionName",
# monitoring_type: "DataQuality", # accepts DataQuality, ModelQuality, ModelBias, ModelExplainability
# },
# })
#
# @example Response structure
#
# resp.monitoring_schedule_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateMonitoringSchedule AWS API Documentation
#
# @overload update_monitoring_schedule(params = {})
# @param [Hash] params ({})
def update_monitoring_schedule(params = {}, options = {})
req = build_request(:update_monitoring_schedule, params)
req.send_request(options)
end
# Updates a notebook instance. NotebookInstance updates include
# upgrading or downgrading the ML compute instance used for your
# notebook instance to accommodate changes in your workload
# requirements.
#
# @option params [required, String] :notebook_instance_name
# The name of the notebook instance to update.
#
# @option params [String] :instance_type
# The Amazon ML compute instance type.
#
# @option params [String] :role_arn
# The Amazon Resource Name (ARN) of the IAM role that SageMaker can
# assume to access the notebook instance. For more information, see
# [SageMaker Roles][1].
#
# To be able to pass this role to SageMaker, the caller of this API must
# have the `iam:PassRole` permission.
#
#
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html
#
# @option params [String] :lifecycle_config_name
# The name of a lifecycle configuration to associate with the notebook
# instance. For information about lifestyle configurations, see [Step
# 2.1: (Optional) Customize a Notebook Instance][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/notebook-lifecycle-config.html
#
# @option params [Boolean] :disassociate_lifecycle_config
# Set to `true` to remove the notebook instance lifecycle configuration
# currently associated with the notebook instance. This operation is
# idempotent. If you specify a lifecycle configuration that is not
# associated with the notebook instance when you call this method, it
# does not throw an error.
#
# @option params [Integer] :volume_size_in_gb
# The size, in GB, of the ML storage volume to attach to the notebook
# instance. The default value is 5 GB. ML storage volumes are encrypted,
# so SageMaker can't determine the amount of available free space on
# the volume. Because of this, you can increase the volume size when you
# update a notebook instance, but you can't decrease the volume size.
# If you want to decrease the size of the ML storage volume in use,
# create a new notebook instance with the desired size.
#
# @option params [String] :default_code_repository
# The Git repository to associate with the notebook instance as its
# default code repository. This can be either the name of a Git
# repository stored as a resource in your account, or the URL of a Git
# repository in [Amazon Web Services CodeCommit][1] or in any other Git
# repository. When you open a notebook instance, it opens in the
# directory that contains this repository. For more information, see
# [Associating Git Repositories with SageMaker Notebook Instances][2].
#
#
#
# [1]: https://docs.aws.amazon.com/codecommit/latest/userguide/welcome.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/nbi-git-repo.html
#
# @option params [Array] :additional_code_repositories
# An array of up to three Git repositories to associate with the
# notebook instance. These can be either the names of Git repositories
# stored as resources in your account, or the URL of Git repositories in
# [Amazon Web Services CodeCommit][1] or in any other Git repository.
# These repositories are cloned at the same level as the default
# repository of your notebook instance. For more information, see
# [Associating Git Repositories with SageMaker Notebook Instances][2].
#
#
#
# [1]: https://docs.aws.amazon.com/codecommit/latest/userguide/welcome.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/dg/nbi-git-repo.html
#
# @option params [Array] :accelerator_types
# A list of the Elastic Inference (EI) instance types to associate with
# this notebook instance. Currently only one EI instance type can be
# associated with a notebook instance. For more information, see [Using
# Elastic Inference in Amazon SageMaker][1].
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/dg/ei.html
#
# @option params [Boolean] :disassociate_accelerator_types
# A list of the Elastic Inference (EI) instance types to remove from
# this notebook instance. This operation is idempotent. If you specify
# an accelerator type that is not associated with the notebook instance
# when you call this method, it does not throw an error.
#
# @option params [Boolean] :disassociate_default_code_repository
# The name or URL of the default Git repository to remove from this
# notebook instance. This operation is idempotent. If you specify a Git
# repository that is not associated with the notebook instance when you
# call this method, it does not throw an error.
#
# @option params [Boolean] :disassociate_additional_code_repositories
# A list of names or URLs of the default Git repositories to remove from
# this notebook instance. This operation is idempotent. If you specify a
# Git repository that is not associated with the notebook instance when
# you call this method, it does not throw an error.
#
# @option params [String] :root_access
# Whether root access is enabled or disabled for users of the notebook
# instance. The default value is `Enabled`.
#
# If you set this to `Disabled`, users don't have root access on the
# notebook instance, but lifecycle configuration scripts still run with
# root permissions.
#
#
#
# @option params [Types::InstanceMetadataServiceConfiguration] :instance_metadata_service_configuration
# Information on the IMDS configuration of the notebook instance
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.update_notebook_instance({
# notebook_instance_name: "NotebookInstanceName", # required
# instance_type: "ml.t2.medium", # accepts ml.t2.medium, ml.t2.large, ml.t2.xlarge, ml.t2.2xlarge, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.c5d.xlarge, ml.c5d.2xlarge, ml.c5d.4xlarge, ml.c5d.9xlarge, ml.c5d.18xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge
# role_arn: "RoleArn",
# lifecycle_config_name: "NotebookInstanceLifecycleConfigName",
# disassociate_lifecycle_config: false,
# volume_size_in_gb: 1,
# default_code_repository: "CodeRepositoryNameOrUrl",
# additional_code_repositories: ["CodeRepositoryNameOrUrl"],
# accelerator_types: ["ml.eia1.medium"], # accepts ml.eia1.medium, ml.eia1.large, ml.eia1.xlarge, ml.eia2.medium, ml.eia2.large, ml.eia2.xlarge
# disassociate_accelerator_types: false,
# disassociate_default_code_repository: false,
# disassociate_additional_code_repositories: false,
# root_access: "Enabled", # accepts Enabled, Disabled
# instance_metadata_service_configuration: {
# minimum_instance_metadata_service_version: "MinimumInstanceMetadataServiceVersion", # required
# },
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateNotebookInstance AWS API Documentation
#
# @overload update_notebook_instance(params = {})
# @param [Hash] params ({})
def update_notebook_instance(params = {}, options = {})
req = build_request(:update_notebook_instance, params)
req.send_request(options)
end
# Updates a notebook instance lifecycle configuration created with the
# [CreateNotebookInstanceLifecycleConfig][1] API.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateNotebookInstanceLifecycleConfig.html
#
# @option params [required, String] :notebook_instance_lifecycle_config_name
# The name of the lifecycle configuration.
#
# @option params [Array] :on_create
# The shell script that runs only once, when you create a notebook
# instance. The shell script must be a base64-encoded string.
#
# @option params [Array] :on_start
# The shell script that runs every time you start a notebook instance,
# including when you create the notebook instance. The shell script must
# be a base64-encoded string.
#
# @return [Struct] Returns an empty {Seahorse::Client::Response response}.
#
# @example Request syntax with placeholder values
#
# resp = client.update_notebook_instance_lifecycle_config({
# notebook_instance_lifecycle_config_name: "NotebookInstanceLifecycleConfigName", # required
# on_create: [
# {
# content: "NotebookInstanceLifecycleConfigContent",
# },
# ],
# on_start: [
# {
# content: "NotebookInstanceLifecycleConfigContent",
# },
# ],
# })
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateNotebookInstanceLifecycleConfig AWS API Documentation
#
# @overload update_notebook_instance_lifecycle_config(params = {})
# @param [Hash] params ({})
def update_notebook_instance_lifecycle_config(params = {}, options = {})
req = build_request(:update_notebook_instance_lifecycle_config, params)
req.send_request(options)
end
# Updates a pipeline.
#
# @option params [required, String] :pipeline_name
# The name of the pipeline to update.
#
# @option params [String] :pipeline_display_name
# The display name of the pipeline.
#
# @option params [String] :pipeline_definition
# The JSON pipeline definition.
#
# @option params [Types::PipelineDefinitionS3Location] :pipeline_definition_s3_location
# The location of the pipeline definition stored in Amazon S3. If
# specified, SageMaker will retrieve the pipeline definition from this
# location.
#
# @option params [String] :pipeline_description
# The description of the pipeline.
#
# @option params [String] :role_arn
# The Amazon Resource Name (ARN) that the pipeline uses to execute.
#
# @option params [Types::ParallelismConfiguration] :parallelism_configuration
# If specified, it applies to all executions of this pipeline by
# default.
#
# @return [Types::UpdatePipelineResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdatePipelineResponse#pipeline_arn #pipeline_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_pipeline({
# pipeline_name: "PipelineName", # required
# pipeline_display_name: "PipelineName",
# pipeline_definition: "PipelineDefinition",
# pipeline_definition_s3_location: {
# bucket: "BucketName", # required
# object_key: "Key", # required
# version_id: "VersionId",
# },
# pipeline_description: "PipelineDescription",
# role_arn: "RoleArn",
# parallelism_configuration: {
# max_parallel_execution_steps: 1, # required
# },
# })
#
# @example Response structure
#
# resp.pipeline_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdatePipeline AWS API Documentation
#
# @overload update_pipeline(params = {})
# @param [Hash] params ({})
def update_pipeline(params = {}, options = {})
req = build_request(:update_pipeline, params)
req.send_request(options)
end
# Updates a pipeline execution.
#
# @option params [required, String] :pipeline_execution_arn
# The Amazon Resource Name (ARN) of the pipeline execution.
#
# @option params [String] :pipeline_execution_description
# The description of the pipeline execution.
#
# @option params [String] :pipeline_execution_display_name
# The display name of the pipeline execution.
#
# @option params [Types::ParallelismConfiguration] :parallelism_configuration
# This configuration, if specified, overrides the parallelism
# configuration of the parent pipeline for this specific run.
#
# @return [Types::UpdatePipelineExecutionResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdatePipelineExecutionResponse#pipeline_execution_arn #pipeline_execution_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_pipeline_execution({
# pipeline_execution_arn: "PipelineExecutionArn", # required
# pipeline_execution_description: "PipelineExecutionDescription",
# pipeline_execution_display_name: "PipelineExecutionName",
# parallelism_configuration: {
# max_parallel_execution_steps: 1, # required
# },
# })
#
# @example Response structure
#
# resp.pipeline_execution_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdatePipelineExecution AWS API Documentation
#
# @overload update_pipeline_execution(params = {})
# @param [Hash] params ({})
def update_pipeline_execution(params = {}, options = {})
req = build_request(:update_pipeline_execution, params)
req.send_request(options)
end
# Updates a machine learning (ML) project that is created from a
# template that sets up an ML pipeline from training to deploying an
# approved model.
#
# You must not update a project that is in use. If you update the
# `ServiceCatalogProvisioningUpdateDetails` of a project that is active
# or being created, or updated, you may lose resources already created
# by the project.
#
#
#
# @option params [required, String] :project_name
# The name of the project.
#
# @option params [String] :project_description
# The description for the project.
#
# @option params [Types::ServiceCatalogProvisioningUpdateDetails] :service_catalog_provisioning_update_details
# The product ID and provisioning artifact ID to provision a service
# catalog. The provisioning artifact ID will default to the latest
# provisioning artifact ID of the product, if you don't provide the
# provisioning artifact ID. For more information, see [What is Amazon
# Web Services Service Catalog][1].
#
#
#
# [1]: https://docs.aws.amazon.com/servicecatalog/latest/adminguide/introduction.html
#
# @option params [Array] :tags
# An array of key-value pairs. You can use tags to categorize your
# Amazon Web Services resources in different ways, for example, by
# purpose, owner, or environment. For more information, see [Tagging
# Amazon Web Services Resources][1]. In addition, the project must have
# tag update constraints set in order to include this parameter in the
# request. For more information, see [Amazon Web Services Service
# Catalog Tag Update Constraints][2].
#
#
#
# [1]: https://docs.aws.amazon.com/general/latest/gr/aws_tagging.html
# [2]: https://docs.aws.amazon.com/servicecatalog/latest/adminguide/constraints-resourceupdate.html
#
# @return [Types::UpdateProjectOutput] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateProjectOutput#project_arn #project_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_project({
# project_name: "ProjectEntityName", # required
# project_description: "EntityDescription",
# service_catalog_provisioning_update_details: {
# provisioning_artifact_id: "ServiceCatalogEntityId",
# provisioning_parameters: [
# {
# key: "ProvisioningParameterKey",
# value: "ProvisioningParameterValue",
# },
# ],
# },
# tags: [
# {
# key: "TagKey", # required
# value: "TagValue", # required
# },
# ],
# })
#
# @example Response structure
#
# resp.project_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateProject AWS API Documentation
#
# @overload update_project(params = {})
# @param [Hash] params ({})
def update_project(params = {}, options = {})
req = build_request(:update_project, params)
req.send_request(options)
end
# Updates the settings of a space.
#
# @option params [required, String] :domain_id
# The ID of the associated Domain.
#
# @option params [required, String] :space_name
# The name of the space.
#
# @option params [Types::SpaceSettings] :space_settings
# A collection of space settings.
#
# @return [Types::UpdateSpaceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateSpaceResponse#space_arn #space_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_space({
# domain_id: "DomainId", # required
# space_name: "SpaceName", # required
# space_settings: {
# jupyter_server_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# code_repositories: [
# {
# repository_url: "RepositoryUrl", # required
# },
# ],
# },
# kernel_gateway_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# },
# },
# })
#
# @example Response structure
#
# resp.space_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateSpace AWS API Documentation
#
# @overload update_space(params = {})
# @param [Hash] params ({})
def update_space(params = {}, options = {})
req = build_request(:update_space, params)
req.send_request(options)
end
# Update a model training job to request a new Debugger profiling
# configuration or to change warm pool retention length.
#
# @option params [required, String] :training_job_name
# The name of a training job to update the Debugger profiling
# configuration.
#
# @option params [Types::ProfilerConfigForUpdate] :profiler_config
# Configuration information for Amazon SageMaker Debugger system
# monitoring, framework profiling, and storage paths.
#
# @option params [Array] :profiler_rule_configurations
# Configuration information for Amazon SageMaker Debugger rules for
# profiling system and framework metrics.
#
# @option params [Types::ResourceConfigForUpdate] :resource_config
# The training job `ResourceConfig` to update warm pool retention
# length.
#
# @return [Types::UpdateTrainingJobResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateTrainingJobResponse#training_job_arn #training_job_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_training_job({
# training_job_name: "TrainingJobName", # required
# profiler_config: {
# s3_output_path: "S3Uri",
# profiling_interval_in_milliseconds: 1,
# profiling_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# disable_profiler: false,
# },
# profiler_rule_configurations: [
# {
# rule_configuration_name: "RuleConfigurationName", # required
# local_path: "DirectoryPath",
# s3_output_path: "S3Uri",
# rule_evaluator_image: "AlgorithmImage", # required
# instance_type: "ml.t3.medium", # accepts ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m4.xlarge, ml.m4.2xlarge, ml.m4.4xlarge, ml.m4.10xlarge, ml.m4.16xlarge, ml.c4.xlarge, ml.c4.2xlarge, ml.c4.4xlarge, ml.c4.8xlarge, ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.18xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.12xlarge, ml.m5.24xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge
# volume_size_in_gb: 1,
# rule_parameters: {
# "ConfigKey" => "ConfigValue",
# },
# },
# ],
# resource_config: {
# keep_alive_period_in_seconds: 1, # required
# },
# })
#
# @example Response structure
#
# resp.training_job_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateTrainingJob AWS API Documentation
#
# @overload update_training_job(params = {})
# @param [Hash] params ({})
def update_training_job(params = {}, options = {})
req = build_request(:update_training_job, params)
req.send_request(options)
end
# Updates the display name of a trial.
#
# @option params [required, String] :trial_name
# The name of the trial to update.
#
# @option params [String] :display_name
# The name of the trial as displayed. The name doesn't need to be
# unique. If `DisplayName` isn't specified, `TrialName` is displayed.
#
# @return [Types::UpdateTrialResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateTrialResponse#trial_arn #trial_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_trial({
# trial_name: "ExperimentEntityName", # required
# display_name: "ExperimentEntityName",
# })
#
# @example Response structure
#
# resp.trial_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateTrial AWS API Documentation
#
# @overload update_trial(params = {})
# @param [Hash] params ({})
def update_trial(params = {}, options = {})
req = build_request(:update_trial, params)
req.send_request(options)
end
# Updates one or more properties of a trial component.
#
# @option params [required, String] :trial_component_name
# The name of the component to update.
#
# @option params [String] :display_name
# The name of the component as displayed. The name doesn't need to be
# unique. If `DisplayName` isn't specified, `TrialComponentName` is
# displayed.
#
# @option params [Types::TrialComponentStatus] :status
# The new status of the component.
#
# @option params [Time,DateTime,Date,Integer,String] :start_time
# When the component started.
#
# @option params [Time,DateTime,Date,Integer,String] :end_time
# When the component ended.
#
# @option params [Hash] :parameters
# Replaces all of the component's hyperparameters with the specified
# hyperparameters or add new hyperparameters. Existing hyperparameters
# are replaced if the trial component is updated with an identical
# hyperparameter key.
#
# @option params [Array] :parameters_to_remove
# The hyperparameters to remove from the component.
#
# @option params [Hash] :input_artifacts
# Replaces all of the component's input artifacts with the specified
# artifacts or adds new input artifacts. Existing input artifacts are
# replaced if the trial component is updated with an identical input
# artifact key.
#
# @option params [Array] :input_artifacts_to_remove
# The input artifacts to remove from the component.
#
# @option params [Hash] :output_artifacts
# Replaces all of the component's output artifacts with the specified
# artifacts or adds new output artifacts. Existing output artifacts are
# replaced if the trial component is updated with an identical output
# artifact key.
#
# @option params [Array] :output_artifacts_to_remove
# The output artifacts to remove from the component.
#
# @return [Types::UpdateTrialComponentResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateTrialComponentResponse#trial_component_arn #trial_component_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_trial_component({
# trial_component_name: "ExperimentEntityName", # required
# display_name: "ExperimentEntityName",
# status: {
# primary_status: "InProgress", # accepts InProgress, Completed, Failed, Stopping, Stopped
# message: "TrialComponentStatusMessage",
# },
# start_time: Time.now,
# end_time: Time.now,
# parameters: {
# "TrialComponentKey256" => {
# string_value: "StringParameterValue",
# number_value: 1.0,
# },
# },
# parameters_to_remove: ["TrialComponentKey256"],
# input_artifacts: {
# "TrialComponentKey64" => {
# media_type: "MediaType",
# value: "TrialComponentArtifactValue", # required
# },
# },
# input_artifacts_to_remove: ["TrialComponentKey256"],
# output_artifacts: {
# "TrialComponentKey64" => {
# media_type: "MediaType",
# value: "TrialComponentArtifactValue", # required
# },
# },
# output_artifacts_to_remove: ["TrialComponentKey256"],
# })
#
# @example Response structure
#
# resp.trial_component_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateTrialComponent AWS API Documentation
#
# @overload update_trial_component(params = {})
# @param [Hash] params ({})
def update_trial_component(params = {}, options = {})
req = build_request(:update_trial_component, params)
req.send_request(options)
end
# Updates a user profile.
#
# @option params [required, String] :domain_id
# The domain ID.
#
# @option params [required, String] :user_profile_name
# The user profile name.
#
# @option params [Types::UserSettings] :user_settings
# A collection of settings.
#
# @return [Types::UpdateUserProfileResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateUserProfileResponse#user_profile_arn #user_profile_arn} => String
#
# @example Request syntax with placeholder values
#
# resp = client.update_user_profile({
# domain_id: "DomainId", # required
# user_profile_name: "UserProfileName", # required
# user_settings: {
# execution_role: "RoleArn",
# security_groups: ["SecurityGroupId"],
# sharing_settings: {
# notebook_output_option: "Allowed", # accepts Allowed, Disabled
# s3_output_path: "S3Uri",
# s3_kms_key_id: "KmsKeyId",
# },
# jupyter_server_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# code_repositories: [
# {
# repository_url: "RepositoryUrl", # required
# },
# ],
# },
# kernel_gateway_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# lifecycle_config_arns: ["StudioLifecycleConfigArn"],
# },
# tensor_board_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# },
# r_studio_server_pro_app_settings: {
# access_status: "ENABLED", # accepts ENABLED, DISABLED
# user_group: "R_STUDIO_ADMIN", # accepts R_STUDIO_ADMIN, R_STUDIO_USER
# },
# r_session_app_settings: {
# default_resource_spec: {
# sage_maker_image_arn: "ImageArn",
# sage_maker_image_version_arn: "ImageVersionArn",
# instance_type: "system", # accepts system, ml.t3.micro, ml.t3.small, ml.t3.medium, ml.t3.large, ml.t3.xlarge, ml.t3.2xlarge, ml.m5.large, ml.m5.xlarge, ml.m5.2xlarge, ml.m5.4xlarge, ml.m5.8xlarge, ml.m5.12xlarge, ml.m5.16xlarge, ml.m5.24xlarge, ml.m5d.large, ml.m5d.xlarge, ml.m5d.2xlarge, ml.m5d.4xlarge, ml.m5d.8xlarge, ml.m5d.12xlarge, ml.m5d.16xlarge, ml.m5d.24xlarge, ml.c5.large, ml.c5.xlarge, ml.c5.2xlarge, ml.c5.4xlarge, ml.c5.9xlarge, ml.c5.12xlarge, ml.c5.18xlarge, ml.c5.24xlarge, ml.p3.2xlarge, ml.p3.8xlarge, ml.p3.16xlarge, ml.p3dn.24xlarge, ml.g4dn.xlarge, ml.g4dn.2xlarge, ml.g4dn.4xlarge, ml.g4dn.8xlarge, ml.g4dn.12xlarge, ml.g4dn.16xlarge, ml.r5.large, ml.r5.xlarge, ml.r5.2xlarge, ml.r5.4xlarge, ml.r5.8xlarge, ml.r5.12xlarge, ml.r5.16xlarge, ml.r5.24xlarge, ml.g5.xlarge, ml.g5.2xlarge, ml.g5.4xlarge, ml.g5.8xlarge, ml.g5.16xlarge, ml.g5.12xlarge, ml.g5.24xlarge, ml.g5.48xlarge, ml.geospatial.interactive, ml.p4d.24xlarge, ml.p4de.24xlarge
# lifecycle_config_arn: "StudioLifecycleConfigArn",
# },
# custom_images: [
# {
# image_name: "ImageName", # required
# image_version_number: 1,
# app_image_config_name: "AppImageConfigName", # required
# },
# ],
# },
# canvas_app_settings: {
# time_series_forecasting_settings: {
# status: "ENABLED", # accepts ENABLED, DISABLED
# amazon_forecast_role_arn: "RoleArn",
# },
# model_register_settings: {
# status: "ENABLED", # accepts ENABLED, DISABLED
# cross_account_model_register_role_arn: "RoleArn",
# },
# },
# },
# })
#
# @example Response structure
#
# resp.user_profile_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateUserProfile AWS API Documentation
#
# @overload update_user_profile(params = {})
# @param [Hash] params ({})
def update_user_profile(params = {}, options = {})
req = build_request(:update_user_profile, params)
req.send_request(options)
end
# Use this operation to update your workforce. You can use this
# operation to require that workers use specific IP addresses to work on
# tasks and to update your OpenID Connect (OIDC) Identity Provider (IdP)
# workforce configuration.
#
# The worker portal is now supported in VPC and public internet.
#
# Use `SourceIpConfig` to restrict worker access to tasks to a specific
# range of IP addresses. You specify allowed IP addresses by creating a
# list of up to ten [CIDRs][1]. By default, a workforce isn't
# restricted to specific IP addresses. If you specify a range of IP
# addresses, workers who attempt to access tasks using any IP address
# outside the specified range are denied and get a `Not Found` error
# message on the worker portal.
#
# To restrict access to all the workers in public internet, add the
# `SourceIpConfig` CIDR value as "10.0.0.0/16".
#
# Amazon SageMaker does not support Source Ip restriction for worker
# portals in VPC.
#
# Use `OidcConfig` to update the configuration of a workforce created
# using your own OIDC IdP.
#
# You can only update your OIDC IdP configuration when there are no work
# teams associated with your workforce. You can delete work teams using
# the [DeleteWorkteam][2] operation.
#
# After restricting access to a range of IP addresses or updating your
# OIDC IdP configuration with this operation, you can view details about
# your update workforce using the [DescribeWorkforce][3] operation.
#
# This operation only applies to private workforces.
#
#
#
# [1]: https://docs.aws.amazon.com/vpc/latest/userguide/VPC_Subnets.html
# [2]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DeleteWorkteam.html
# [3]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeWorkforce.html
#
# @option params [required, String] :workforce_name
# The name of the private workforce that you want to update. You can
# find your workforce name by using the [ListWorkforces][1] operation.
#
#
#
# [1]: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ListWorkforces.html
#
# @option params [Types::SourceIpConfig] :source_ip_config
# A list of one to ten worker IP address ranges ([CIDRs][1]) that can be
# used to access tasks assigned to this workforce.
#
# Maximum: Ten CIDR values
#
#
#
# [1]: https://docs.aws.amazon.com/vpc/latest/userguide/VPC_Subnets.html
#
# @option params [Types::OidcConfig] :oidc_config
# Use this parameter to update your OIDC Identity Provider (IdP)
# configuration for a workforce made using your own IdP.
#
# @option params [Types::WorkforceVpcConfigRequest] :workforce_vpc_config
# Use this parameter to update your VPC configuration for a workforce.
#
# @return [Types::UpdateWorkforceResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateWorkforceResponse#workforce #workforce} => Types::Workforce
#
# @example Request syntax with placeholder values
#
# resp = client.update_workforce({
# workforce_name: "WorkforceName", # required
# source_ip_config: {
# cidrs: ["Cidr"], # required
# },
# oidc_config: {
# client_id: "ClientId", # required
# client_secret: "ClientSecret", # required
# issuer: "OidcEndpoint", # required
# authorization_endpoint: "OidcEndpoint", # required
# token_endpoint: "OidcEndpoint", # required
# user_info_endpoint: "OidcEndpoint", # required
# logout_endpoint: "OidcEndpoint", # required
# jwks_uri: "OidcEndpoint", # required
# },
# workforce_vpc_config: {
# vpc_id: "WorkforceVpcId",
# security_group_ids: ["WorkforceSecurityGroupId"],
# subnets: ["WorkforceSubnetId"],
# },
# })
#
# @example Response structure
#
# resp.workforce.workforce_name #=> String
# resp.workforce.workforce_arn #=> String
# resp.workforce.last_updated_date #=> Time
# resp.workforce.source_ip_config.cidrs #=> Array
# resp.workforce.source_ip_config.cidrs[0] #=> String
# resp.workforce.sub_domain #=> String
# resp.workforce.cognito_config.user_pool #=> String
# resp.workforce.cognito_config.client_id #=> String
# resp.workforce.oidc_config.client_id #=> String
# resp.workforce.oidc_config.issuer #=> String
# resp.workforce.oidc_config.authorization_endpoint #=> String
# resp.workforce.oidc_config.token_endpoint #=> String
# resp.workforce.oidc_config.user_info_endpoint #=> String
# resp.workforce.oidc_config.logout_endpoint #=> String
# resp.workforce.oidc_config.jwks_uri #=> String
# resp.workforce.create_date #=> Time
# resp.workforce.workforce_vpc_config.vpc_id #=> String
# resp.workforce.workforce_vpc_config.security_group_ids #=> Array
# resp.workforce.workforce_vpc_config.security_group_ids[0] #=> String
# resp.workforce.workforce_vpc_config.subnets #=> Array
# resp.workforce.workforce_vpc_config.subnets[0] #=> String
# resp.workforce.workforce_vpc_config.vpc_endpoint_id #=> String
# resp.workforce.status #=> String, one of "Initializing", "Updating", "Deleting", "Failed", "Active"
# resp.workforce.failure_reason #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateWorkforce AWS API Documentation
#
# @overload update_workforce(params = {})
# @param [Hash] params ({})
def update_workforce(params = {}, options = {})
req = build_request(:update_workforce, params)
req.send_request(options)
end
# Updates an existing work team with new member definitions or
# description.
#
# @option params [required, String] :workteam_name
# The name of the work team to update.
#
# @option params [Array] :member_definitions
# A list of `MemberDefinition` objects that contains objects that
# identify the workers that make up the work team.
#
# Workforces can be created using Amazon Cognito or your own OIDC
# Identity Provider (IdP). For private workforces created using Amazon
# Cognito use `CognitoMemberDefinition`. For workforces created using
# your own OIDC identity provider (IdP) use `OidcMemberDefinition`. You
# should not provide input for both of these parameters in a single
# request.
#
# For workforces created using Amazon Cognito, private work teams
# correspond to Amazon Cognito *user groups* within the user pool used
# to create a workforce. All of the `CognitoMemberDefinition` objects
# that make up the member definition must have the same `ClientId` and
# `UserPool` values. To add a Amazon Cognito user group to an existing
# worker pool, see [Adding groups to a User Pool](). For more
# information about user pools, see [Amazon Cognito User Pools][1].
#
# For workforces created using your own OIDC IdP, specify the user
# groups that you want to include in your private work team in
# `OidcMemberDefinition` by listing those groups in `Groups`. Be aware
# that user groups that are already in the work team must also be listed
# in `Groups` when you make this request to remain on the work team. If
# you do not include these user groups, they will no longer be
# associated with the work team you update.
#
#
#
# [1]: https://docs.aws.amazon.com/cognito/latest/developerguide/cognito-user-identity-pools.html
#
# @option params [String] :description
# An updated description for the work team.
#
# @option params [Types::NotificationConfiguration] :notification_configuration
# Configures SNS topic notifications for available or expiring work
# items
#
# @return [Types::UpdateWorkteamResponse] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
#
# * {Types::UpdateWorkteamResponse#workteam #workteam} => Types::Workteam
#
# @example Request syntax with placeholder values
#
# resp = client.update_workteam({
# workteam_name: "WorkteamName", # required
# member_definitions: [
# {
# cognito_member_definition: {
# user_pool: "CognitoUserPool", # required
# user_group: "CognitoUserGroup", # required
# client_id: "ClientId", # required
# },
# oidc_member_definition: {
# groups: ["Group"], # required
# },
# },
# ],
# description: "String200",
# notification_configuration: {
# notification_topic_arn: "NotificationTopicArn",
# },
# })
#
# @example Response structure
#
# resp.workteam.workteam_name #=> String
# resp.workteam.member_definitions #=> Array
# resp.workteam.member_definitions[0].cognito_member_definition.user_pool #=> String
# resp.workteam.member_definitions[0].cognito_member_definition.user_group #=> String
# resp.workteam.member_definitions[0].cognito_member_definition.client_id #=> String
# resp.workteam.member_definitions[0].oidc_member_definition.groups #=> Array
# resp.workteam.member_definitions[0].oidc_member_definition.groups[0] #=> String
# resp.workteam.workteam_arn #=> String
# resp.workteam.workforce_arn #=> String
# resp.workteam.product_listing_ids #=> Array
# resp.workteam.product_listing_ids[0] #=> String
# resp.workteam.description #=> String
# resp.workteam.sub_domain #=> String
# resp.workteam.create_date #=> Time
# resp.workteam.last_updated_date #=> Time
# resp.workteam.notification_configuration.notification_topic_arn #=> String
#
# @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/UpdateWorkteam AWS API Documentation
#
# @overload update_workteam(params = {})
# @param [Hash] params ({})
def update_workteam(params = {}, options = {})
req = build_request(:update_workteam, params)
req.send_request(options)
end
# @!endgroup
# @param params ({})
# @api private
def build_request(operation_name, params = {})
handlers = @handlers.for(operation_name)
context = Seahorse::Client::RequestContext.new(
operation_name: operation_name,
operation: config.api.operation(operation_name),
client: self,
params: params,
config: config)
context[:gem_name] = 'aws-sdk-sagemaker'
context[:gem_version] = '1.177.0'
Seahorse::Client::Request.new(handlers, context)
end
# Polls an API operation until a resource enters a desired state.
#
# ## Basic Usage
#
# A waiter will call an API operation until:
#
# * It is successful
# * It enters a terminal state
# * It makes the maximum number of attempts
#
# In between attempts, the waiter will sleep.
#
# # polls in a loop, sleeping between attempts
# client.wait_until(waiter_name, params)
#
# ## Configuration
#
# You can configure the maximum number of polling attempts, and the
# delay (in seconds) between each polling attempt. You can pass
# configuration as the final arguments hash.
#
# # poll for ~25 seconds
# client.wait_until(waiter_name, params, {
# max_attempts: 5,
# delay: 5,
# })
#
# ## Callbacks
#
# You can be notified before each polling attempt and before each
# delay. If you throw `:success` or `:failure` from these callbacks,
# it will terminate the waiter.
#
# started_at = Time.now
# client.wait_until(waiter_name, params, {
#
# # disable max attempts
# max_attempts: nil,
#
# # poll for 1 hour, instead of a number of attempts
# before_wait: -> (attempts, response) do
# throw :failure if Time.now - started_at > 3600
# end
# })
#
# ## Handling Errors
#
# When a waiter is unsuccessful, it will raise an error.
# All of the failure errors extend from
# {Aws::Waiters::Errors::WaiterFailed}.
#
# begin
# client.wait_until(...)
# rescue Aws::Waiters::Errors::WaiterFailed
# # resource did not enter the desired state in time
# end
#
# ## Valid Waiters
#
# The following table lists the valid waiter names, the operations they call,
# and the default `:delay` and `:max_attempts` values.
#
# | waiter_name | params | :delay | :max_attempts |
# | ----------------------------------- | ----------------------------------- | -------- | ------------- |
# | endpoint_deleted | {Client#describe_endpoint} | 30 | 60 |
# | endpoint_in_service | {Client#describe_endpoint} | 30 | 120 |
# | image_created | {Client#describe_image} | 60 | 60 |
# | image_deleted | {Client#describe_image} | 60 | 60 |
# | image_updated | {Client#describe_image} | 60 | 60 |
# | image_version_created | {Client#describe_image_version} | 60 | 60 |
# | image_version_deleted | {Client#describe_image_version} | 60 | 60 |
# | notebook_instance_deleted | {Client#describe_notebook_instance} | 30 | 60 |
# | notebook_instance_in_service | {Client#describe_notebook_instance} | 30 | 60 |
# | notebook_instance_stopped | {Client#describe_notebook_instance} | 30 | 60 |
# | processing_job_completed_or_stopped | {Client#describe_processing_job} | 60 | 60 |
# | training_job_completed_or_stopped | {Client#describe_training_job} | 120 | 180 |
# | transform_job_completed_or_stopped | {Client#describe_transform_job} | 60 | 60 |
#
# @raise [Errors::FailureStateError] Raised when the waiter terminates
# because the waiter has entered a state that it will not transition
# out of, preventing success.
#
# @raise [Errors::TooManyAttemptsError] Raised when the configured
# maximum number of attempts have been made, and the waiter is not
# yet successful.
#
# @raise [Errors::UnexpectedError] Raised when an error is encounted
# while polling for a resource that is not expected.
#
# @raise [Errors::NoSuchWaiterError] Raised when you request to wait
# for an unknown state.
#
# @return [Boolean] Returns `true` if the waiter was successful.
# @param [Symbol] waiter_name
# @param [Hash] params ({})
# @param [Hash] options ({})
# @option options [Integer] :max_attempts
# @option options [Integer] :delay
# @option options [Proc] :before_attempt
# @option options [Proc] :before_wait
def wait_until(waiter_name, params = {}, options = {})
w = waiter(waiter_name, options)
yield(w.waiter) if block_given? # deprecated
w.wait(params)
end
# @api private
# @deprecated
def waiter_names
waiters.keys
end
private
# @param [Symbol] waiter_name
# @param [Hash] options ({})
def waiter(waiter_name, options = {})
waiter_class = waiters[waiter_name]
if waiter_class
waiter_class.new(options.merge(client: self))
else
raise Aws::Waiters::Errors::NoSuchWaiterError.new(waiter_name, waiters.keys)
end
end
def waiters
{
endpoint_deleted: Waiters::EndpointDeleted,
endpoint_in_service: Waiters::EndpointInService,
image_created: Waiters::ImageCreated,
image_deleted: Waiters::ImageDeleted,
image_updated: Waiters::ImageUpdated,
image_version_created: Waiters::ImageVersionCreated,
image_version_deleted: Waiters::ImageVersionDeleted,
notebook_instance_deleted: Waiters::NotebookInstanceDeleted,
notebook_instance_in_service: Waiters::NotebookInstanceInService,
notebook_instance_stopped: Waiters::NotebookInstanceStopped,
processing_job_completed_or_stopped: Waiters::ProcessingJobCompletedOrStopped,
training_job_completed_or_stopped: Waiters::TrainingJobCompletedOrStopped,
transform_job_completed_or_stopped: Waiters::TransformJobCompletedOrStopped
}
end
class << self
# @api private
attr_reader :identifier
# @api private
def errors_module
Errors
end
end
end
end