Package

[meilame-tayebjee]

• Fixed issues when no categorical variables
• Completed docs
• Optional dependencies with poetry.extras + smooth imports when necessary + removed calls to numpy when there is an equivalent in torch. "pytorch_model.py" now does not need numpy.
• Added build_data_loaders (is useful for training externally, e.g. codif APE)
• Debugged predict and predict_and_explain + extended the notebook experiments to include full pipeline until explainability

--> Total independance from numpy might be possible but requires more work ; and might be not relevant anymore (the pytorch model can be distributed without numpy, this is the most important - only optional features are now Numpy-dependent)
--> Ready to release v0.0.0 on PyPI ?

[micedre]

Great, some minor modification on README, and good to publish for me