v2.3.0

Minor Enhancements

• Separate tokenizer from hasher, allowing custom tokenizers. (#162)
• Improved handling of Zero division and Zero vector normalization. (#173)
• Support Numo Gem for performing SVD (#198)

Development Fixes

• remove deprecated has_rdoc in gemspec
• auto-gen-config for Rubocop
• auto-correct offenses
• Freeze all Ruby string literals (#190)
• Migrate TravisCI to GitHub Actions & Update Tested Ruby Versions (#195)
• Test Native and GSL Implementations (#196)
• Actually Use GSL in CI Tests (#197)

Documentation

• Updated Bayes docs for filters (#191)

2.2.0

Major Enhancements

• JRuby Support, thanks to @mach-kernel (#168)
• Add support to reset trained classifiers to their initial state (#143)
• Classifier evaluation and validation (#142)
• Abbility to add custom stopwords at classifier initialization (#129)
• Don't train/untrain the Bayesian classifier with empty word hashes (#132)
• Enable auto categorization if no initial categories (#128)
• Bayes integration test of Memory and Redis backends with real data (#92)
• Memory and Redis backend support (#84)

Minor Enhancements

• improved turkish stopwords (#159)
• Set Redis keys only if they don't exist (#156)
• Require bayes_redis_backend (#157)
• Validation documentation improvements (#150)
• Updated Dokcer image to Ruby 2.4 (#149)
• Classifier validation user documenation (#145)
• Fixed persistance for BayesMemoryBackend (#147)
• Fixed error on requiring 'classifier-reborn' without using Redis (#146)
• Removed magic train untrain methods from docs, (#141)
• Links corrected to point to the new domain (#139)
• Minor docs improvements (#138)
• Return the status of the training/untraining when run (#137)
• Refactoring of backend tests to move duplicate login in the common file (#134)
• Deal with Infinity score in test (#133)
• README file cleaned up to point to the documentation site (#121)
• Added and corrected RDoc for ceratin classes and methods (#122)
• Added favicon link and forced display (#120)
• Updated the truncated LICENSE file (#116)
• Docs visual improvement and refactoring (#119)
• Fixed relative URL issue on nav links and added benchmark data (#118)
• Added custom layout with navigation (#117)
• Created a static site for documentation (#115)
• Removed redis gem from Dockerfile as it is added in gemspec (#113)
• Speed up Docker image rebilding (#112)
• Improved Docker based development documentation (#106)
• Benchmark refactoring, improving efficiency, enhanced reporting (#107)
• Add Vietnamese stopwords (#110)
• Added stop words for Arabic, Bengali, Chinese, Hindi, and Russian (#105)
• Dockerfile and documentation (#104)
• Remove hard dep on Redis and update bin (#96)
• Documented Redis backend performance (#103)
• Rename Bayes memory test class (#102)
• Added Bayes backend benchmarks (#98)
• Disabled Redis disc persistence and refactored integration test (#97)
• Removed useless intermediate variables (#90)

v2.1.0

Major Enhancements

• Fix breaking changes in LSI api. Displays errors instead of raising where possible. #87

v2.0.5

Major Enhancements

• Stopwords get encoded to utf8 (#83)
• Fix searching issues where no document is added to lsi (#77)
• Added method to add custom path to user-created stopword directory (#73)

Minor Enhancements

• Test newer rubies (#85)
• Fixed errors in README (#68, #79, #80)
• Added an option to the bayesian classifier to disable word stemming (#61)
• Added missing parens and renamed some variables (#59)

v2.0.4

Major Enhancements

• Classification thresholds can be enabled or disabled. The default is disabled. The threshold value can be set at initialization time or dynamically during processing (#47)
• Made auto-categorization optional, defaulting to false (#45)
• Added the ability to handle an array of classifications to the constructor (#44)
• Classification with a threshold has been added to the api (#39)

Minor Enhancements

• Documentation around threshold usage (#54)
• Fixed UTF-8 encoding for hasher.rb (#50)
• Removed some unnecessary methods (#43)
• Add optional CachedContentNode (GSL only) (#43)
• Caches the transposed search_vector (#43)
• Added custom marshal_ methods to not save the cache when dumping/loading (#43)
• Optimized some numeric comparisons and iterators (#43)
• Added cached calculation table when computing raw_vectors (#43)
• If a category name is already a symbol, just return it (#45)
• Various Hash improvements (#45)
• Eliminated several Ruby ⚠️s when run with RUBYOPT="-w" (#38)
• Simple performance improvements for the Hasher process (#41)
• Fixes for broken regex splitting for non-ascii characters and removal of the unused punctuation filter (#41)
• Add multiple language stopwords with customizable stop word paths (#40)

Bug Fixes

• Fixed the bug where adding the same category a second time would clobber the category that was already there (#45)
• Fixed deprecation warning for <=> in ls.rb (#33)
• Remove references to Madeline in the README and replace it with Marshal or Redis (#32)

Development Fixes

• Added development dependency on mini_test and added 2.2 to travis.yml (#36)

v2.0.2

Minor Enhancements

• Remove Array#sum monkey patch in favour of #reduce(0, :+) (#20)
• Cache total word counts per category for speed (#4)

Development Fixes

• Add a test for Bayes#untrain_*. (#21)
• Fix link to rb-gsl gem (#24)
• Add helper scripts per Jekyll convention (#25)

Many thanks to @Ch4s3 for all his work on this release!

v2.0.1

Bug Fixes

• Count total unique words using methods supported by Vector and GSL::Vector (#11)

v2.0.0 -- Classifier: Reborn.

Bug Fixes

• Remove mathn dependency (#8)
• Only perform first order transform if total UNIQUE words is greater than 1 (#3)
• Update LSI#remove_item such that they will work with the @items hash. (#2)

Development Fixes

• Exclude Gemfile.lock in .gitignore (#7)