## Unreleased changes

## Version 0.5.0 -- 2022-02-16

This release has breaking changes (deprecation cleanup and internals rework), a new feature (smart weights!) and is officially tested on Ruby 3.1.0 (it's what I use). I've enabled the setting to require MFA to publish this gem, to help protect those who use it.

- Breaking: remove `String#tokenize` core extension; please use `Groupie.tokenize(string)` instead [#39](https://github.com/Narnach/groupie/pull/39)
- Breaking: due to changed internals, YAML serialized data from 0.4.x will lack some of the new internal caches. I'd suggest loading the old data and adding the words from each group to a new (0.5.x) instance of Groupie. [#40](https://github.com/Narnach/groupie/pull/40)
- Feat: add support for smart default weights, reducing the effect of low data on predictions [#40](https://github.com/Narnach/groupie/pull/40)
- Deps: add Ruby 3.1 to list of tested & supported gems
- Chore: require multi-factor authentication to publish gem updates
- Chore: add Security.md to advertise a security policy
- Style: addressed Lint/AmbiguousOperatorPrecedence
- Dev: bump development dependencies multiple times
- Dev: switch to DepFu to manage development dependencies

## Version 0.4.1 -- 2021-09-08

Non-functional fixes to the CI config and Rubygems.org metadata.

- Fix: correct changelog uri for gem
- CI: fix dependabot config

## Version 0.4.0 -- 2021-09-07

Welcome to 2021, where Ruby version 2.6 is the lowest with official support, Bundler is the default for managing packages and RSpec version 3 is used to test things. This version updates Groupie into this decade.

- Refactor: update Groupie to 2021 standards
- Feat: raise Groupie::Error instead of RuntimeError
- Feat: deprecate String#tokenize in favor of Groupie.tokenize
- Doc: document API of Groupie
- Doc: update readme with examples
- Refactor: reorder Groupie methods by importance
- Refactor: simplify Groupie#classify
- Refactor: reduce complexity of Groupie#unique_words
- Refactor: simplify Groupie#classify\_text

## Version 0.3.0 -- 2010-07-29

Multiple changes to the 'unique words' strategy, hopefully improving the behavior.

- Cache unique words in an instance var to reduce time required to do subsequent lookups
- Sanity spec
- Unique strategy now includes all words except for the global 4th quartile
- Unique strategy changed yet again: only ignore words that occur more than their group's median
- Unique strategy now behaves like sqrt that only checks unique words
- Unique word finder uses less elegant but (hopefully) faster code
- Removed gemspec

## Version 0.2.3 -- 2010-07-29

Add a new 'unique words' strategy, which ignores words that occur in all categories.

- Added 'unique' classification strategy
- Added Group#<< as alias for Group#add
- Updated readme

## Version 0.2.2 -- 2010-07-25

Bugfix for log10 strategy.

- Fixed log10 strategy counting for Groupie.classify

## Version 0.2.1 -- 2010-07-25

Offer multiple ways to weigh word counts in calculating final scores.

- Added sqrt and log word counting strategies

## Version 0.2.0 -- 2010-07-25

Classification can't raise division by zero errors anymore.

- Groupie.classify_text ignores unclassified tokens

## Version 0.1.1 -- 2010-07-25

Swap test framework and tokenization improvements.

- Regenerated gemspec
- Strip quotes from tokens
- Replaced testy tests with rspec

## Version 0.1.0 -- 2010-07-25

The initial release as a gem, after working on this on/off over a year.

- Added gemspec
- Fixed text classification to properly average group scores
- Added test for classifying tokenized html email spam
- Classification of texts is now possible
- Added readme and MIT license
- Test the full html and headers of tokenized emails
- Support infix commas for tokenized strings
- Allow infix dots in tokenized strings
- Strip HTML tags when sanitizing a string
- Classify common words based on tokenized text from spam.la e-mails
- Added String#tokenize
- Ensure a Group will still work when loaded from YAML
- Added test helper file
- Refactored Group to maintain a Hash of words and counts instead of a list of words
- Removed obsolete method
- Added testcase for three groups
- Support multiple examples to add more weight to their grouping
- Renamed tests to reflect intent of content
- Classification now allows for a degree of certainty
- Implemented simple spam check