aitp

AI training pipeline using PyTorch (work in progress)

Supported models

• JEDI-net from https://github.com/jmduarte/JEDInet-code

Features

• 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