Lightsaber version 0.3

Highlights

• upgraded pytorch_lightning to version 1.6.4
• updated test script, api largely backported
• Updated example for LSTM/HistGBT

Changes in API:

• pl_trainer.run_training_with_mlflow:

  • Call Signature updated

    • dataloaders can be now passed directly here instead of in wrapped model
    • instead of instantiated trainer, arguments supported by pytorch lightning Trainer is passed as Namespace/ArgumentParser

  • Return type updated. validation/test input and outputs (e.g. y_val, y_pred) are now numpy not torch tensors