- Added top-k accuracy computation, TensorBoardMetricLogger
- ProgressTracker preserves len of iterable
- ProgressTracker tracks iterations inside each epoch
- Cleaner version of checkpoint saver
- Added tests
- Drop support of python 3.5
- Added to_device and apply_recursively helpers
Added tools for training: metric processors, checkpoint creation, progress tracking, dictionary conversion
- added visualization functions
- Setup tests, docs
- First release on PyPI.