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)