Queue Ray backend

Queue Ray backend, so larger numbers of tasks can be sent to Ray at once

Bugfixes

Fixes to updates in 1.4.4 and 1.4.5

32bit FTRL

32-bit precision version of FTRL, and model finalization for 1/3 memory footprint

Apply decorator

Enables use of Apply as a decorator with simplified API. Adds caching and basic Numba support as options for accelerating Apply minibatch processing

Updates, fixes

Several minor fixes and updates

Python library for distributed AI processing pipelines, using swappable scheduler backends.

Support for several distributed backends: Ray, Dask, Spark. Major refactoring and API change.

Refactored pipeline

Refactored wordbatch into a pipeline of separate transformers compatible with sklearn. Added transformers and a modified feature_union class for parallelizing tasks.

Updates & AVX Optimization

Changes for better sklearn-compatibility with fit() and transform() calls.
Optimization of FM_FTRL with AVX instructions.
Started data_utils for assorted util functions.
Several minor bugs fixed.

Spark Integration

1.2.0

Updates for PyPI 1.2 version

1.1.0

Python quote as example