Sha256: fc2af1b182bbd24c0dae13db369b0d894cce2c7d98536c3cc8a3833b57c94fb1

Contents?: true

Size: 1.99 KB

Versions: 4

Compression:

Stored size: 1.99 KB

Contents

module OpenNLP::Config
  
  NameToClass = {
    categorizer: ['DoccatModel', 'opennlp.tools.doccat'],
    chunker: ['ChunkerModel', 'opennlp.tools.chunker'],
    detokenizer: ['DetokenizationDictionary', 'opennlp.tools.tokenize'],
    name_finder: ['TokenNameFinderModel', 'opennlp.tools.namefind'],
    parser: ['ParserModel', 'opennlp.tools.parser'],
    pos_tagger: ['POSModel', 'opennlp.tools.postag'],
    sentence_detector: ['SentenceModel', 'opennlp.tools.sentdetect'],
    tokenizer: ['TokenizerModel', 'opennlp.tools.tokenize']
  }
  
  ClassToName = {
    'ChunkerME' => :chunker,
    'DictionaryDetokenizer' => :detokenizer,
    'DocumentCategorizerME' => :categorizer,
    'NameFinderME' => :name_finder,
    'POSTaggerME' => :pos_tagger,
    'Parser' => :parser,
    'SentenceDetectorME' => :sentence_detector,
    'TokenizerME' => :tokenizer,
  }

  DefaultModels = {
    chunker: {
      english: 'en-chunker.bin'
    },
    detokenizer: {
      english: 'en-detokenizer.xml'
    },
    # Intentionally left empty.
    # Available for English, Spanish, Dutch.
    name_finder: {},
    parser: {
      english: 'en-parser-chunking.bin'
    },
    pos_tagger: { 
      english: 'en-pos-maxent.bin',
      danish: 'da-pos-maxent.bin',
      german: 'de-pos-maxent.bin',
      dutch: 'nl-pos-maxent.bin',
      portuguese: 'pt-pos-maxent.bin',
      swedish: 'se-pos-maxent.bin'
    },
    sentence_detector: {
      english: 'en-sent.bin',
      german: 'de-sent.bin',
      danish: 'da-sent.bin',
      dutch: 'nl-sent.bin',
      portuguese: 'pt-sent.bin',
      swedish: 'se-sent.bin'
    },
    tokenizer: {
      english: 'en-token.bin',
      danish: 'da-token.bin',
      german: 'de-token.bin',
      dutch: 'nl-token.bin',
      portuguese: 'pt-token.bin',
      swedish: 'se-token.bin'
    }
  }

  # Classes that require a model as first argument to constructor.
  RequiresModel = [
    'SentenceDetectorME', 'NameFinderME', 'DictionaryDetokenizer',
    'TokenizerME', 'ChunkerME', 'POSTaggerME', 'Parser'
  ]

  
end

Version data entries

4 entries across 4 versions & 1 rubygems

Version Path
open-nlp-0.1.5 lib/open-nlp/config.rb
open-nlp-0.1.4 lib/open-nlp/config.rb
open-nlp-0.1.3 lib/open-nlp/config.rb
open-nlp-0.1.2 lib/open-nlp/config.rb