ENH: implement handling of class_weights param

[adriangb]

No description provided.

[adriangb]

It looks like for single-output models, this should be straightforwardish. The main complicating factor is that we're going to have to track this through our encode/decode strategy.

For multi-output models, this is a lot more complicated and may require a custom loss function link. I asked in TF (keras-team/keras#4735 (comment)) if there is an easier solution for multi-output models, but I get the feeling that there is not.

[adriangb]

I'm thinking we'll probably want to do this using https://scikit-learn.org/stable/modules/generated/sklearn.utils.class_weight.compute_sample_weight.html instead of relying on Keras. We'll just apply class_weights first then sample_weights.