- Explicitly display "CLIP" in progress bar when gradient norm is clipped - Do not create parent directory for checkpoint file until save is called - Added parquet writer for pandas dataframes
0.2.9
- Homogenized reader and writer directory handling - Added gradient clipping to Trainer
0.2.8
- Added partial to torch cat
0.2.7
- Refactored helper methods in TrainData
0.2.6
- Added epoch argument to training data call method - Changed return signature of TrainData call method - Refactored PyTorch model trainer - Added cosine learning-rate scheduler
0.2.5
- Improved handling of missing parent directory in writers/savers - Absorbed utility to drop None fields from into TOML writer