Add EarlyStoppingShapRFECV to support catboost model

[oleg-savko]

Catboost has a little bit different api and currently EarlyStoppingShapRFECV throw a error, if I tried to use it:

TypeError: fit() got an unexpected keyword argument 'eval_metric'

[Matgrb]

Thanks for making the issue! It would definitely be good to extend support for CatBoost as well.

This requires edits in probatus/feature_elimination/feature_elimination.py in lines 965-982, to pass the arguments differently in case the model is instance of CatBoost object. Would anyone like to pick this issue up?

[Matgrb]

@operte This might be interesting for you

[operte]

CatBoost has quite a different implementation of early stopping. E.g. it has two options for when to stop: https://catboost.ai/docs/concepts/overfitting-detector.html

The approach we took, which is the one taken by LGBM and XGB is equivalent to the Iter approach on CatBoost.

So I guess we need to decide how many of CatBoost's options we want to support.

[Matgrb]

Let's go for the one we currently support for LGBM and XGB, to be consistent.

[ClaudioSalvatoreArcidiacono]

#175 I am proposing a fix for this issue, feedback is welcome, be kind this is my first contribution :)

[ReinierKoops]

I am closing this issue as it seems to be resolved. Feel free to make a new issue if this is not the case.