v0.7.0

• Widens version requirements for scipy and numpy to make compiling dependencies from source less common
• Substantial improvements to documentation. Now using 'alabaster' theme and simplified examples.
• Adds '--test-prop' param to train_test utility.
• Adds a Dockerfile for building an image that will run an ipython notebook to build a revscoring project
• Adds a trim() function for reducing a feature_list to it's basic 'features' -- a prerequisite for https://github.com/wiki-ai/ores/issues/100
• Adds basic language features for dutch, german and italian

We've also fixed out CI process and added code coverage reports.

v0.6.4

Since 0.5.0:

• Drops mediawiki-utilities in favor of mwapi and mwtypes
• Improved error reporting in api extractor
• Change to configuration -- APIExtractor now requires host instead of url
• Fixes an issue with looking up user info in APIExtractor
• Adds batching to Feature extraction for more speed.
• Silences a utf16 encoding warning in enchant
• Fix travis builds, add coverage to reports
• Adds wheel support
• Adds model_info storage & utility

v0.5.0

This release represents a major backwards incompatibility

Languages as feature sets

see #171

Language-based features are now available as modules of the language itself. This allows features of different languages to be combined.

Otherwise, the codebase is now PEP8 compliant.

v0.4.10

• Fixes #160

v0.4.9

• Extended badwords and adds informal words for Persian language
• Fixes pickling issue with languages (see #159)

v0.4.8

Synchronizes dependency versions with https://github.com/wiki-ai/ores and generalizes both.

v0.4.6

• Improved installation instructions
• Selective language imports (no need to download all the dictionaries anymore)
• Fixes #140, #122 and #121.

v0.4.5

• Adds Vietnamese support
• Fixes #140

v0.4.4

• Adds spanish language utilities
• Adds informal words utility to English and Spanish languages
• Converts English language badwords detection to regex based strategy

v0.4.1

• Improves formatting of train_test results (and implements one vs. rest ROC for multiclass models)
• Adds balance_labels arg to constructor of SVC models
• Fixes #122 -- Math domain error when processing imported revisions (user.age)
• Adds 'indonesian' language (thanks @kenrick95!)