The main part of this release is that fixes were made for parallelization for newer versions of TensorFlow. On top of this, python 2.7 support has been dropped. Also, configurable ports were added as well.
0.5.0
This adds some fixes and performance enhancements to the save/load methods.
0.4.0
This release includes a few things:
* Python 3 support * Tensorflow > 1.7 support
Other than these two features, better use of the TensorFlow graph on training was also implemented for faster results.
0.3.0
0.3.0 includes a few smaller features in Sparkflow, including the ability to supply a callback function for losses, and new utility functions to use a pre-trained tensorflow model for predictions.
0.2.1
This latest release includes a new option to add Tensorflow optimization options. This allows the user to specify more options beyond learning rate to their Tensorflow optimization. The other feature added was a way to shuffle partitions between epochs. These recent changes can be downloaded in PyPi