AI training pipeline using PyTorch (work in progress)
- JEDI-net from https://github.com/jmduarte/JEDInet-code
- Running the pipeline's
train.py
script does everything you need; training, validation, plots etc. - Configure behaviour from a single config file
- Automatically produce various result metrics such as ROC curves, AUC, True Positive & False Positive Rates when training completes
- Automatically save checkpoints when the model improves during training
- Automatically pick the checkpoint with lowest loss before running validation
- Label smoothing