Richard Stromer

Thanks for the feedback. I did give it a try with `epoch=5` and `regularization=10`. The problem with so few epochs is that the holiday weights do not take reliable values....

Found a way to make it work with `epoch=10` using tuned hyperparameters and by reducing the batch size for events with the new artificial dataset. Removing the second set of...

Add some events and holidays which do not have a meaningful relationship, rather turn them to noise and assume regularization should reduce their weights.

@ourownstory PR is ready to be reviewed. I've added holidays and events with regular values (therefore meaningless entries) and extended the test cases to expect low weights for those regular...

@ourownstory all comments resolved as previously discussed, hope it's good to go, looking forward to your feedback.

Had to scale regularization down to make sure the meaningful events keep their weights, seems like the regularization scale scales with the number of events and holidays, I created a...

@ourownstory this is the issue we were talking about right? feel free to assign to me then

@ourownstory I could give it a try and see the effect. Otherwise how about integrating linters like flake8, isort, and possibly black in case we want unified code formatting?

Fixed by commit https://github.com/ourownstory/neural_prophet/commit/534f8029fd852223b1234d2e4036db0cef02bb90

Happy to help contribute to this issue as discussed. I've traced down the implementation to these functions: 1) https://github.com/ourownstory/neural_prophet/blob/bae43ab9f934eee81d79fb773ae2852604b81a54/neuralprophet/forecaster.py#L2025-L2028 2) https://github.com/ourownstory/neural_prophet/blob/bae43ab9f934eee81d79fb773ae2852604b81a54/neuralprophet/utils.py#L58 Documentation is currently lacking in NeuralProphet, the [documentation from...