a follow up question: what is the expectation if currently some of the estimators are coded to return pandas dataframes. As an example, in regression.mapie, the _predict function turns y_pred into a pandas Dataframe and returns that as an output.

Should we approach writing this test file under the assumption that all estimators have the capability/functionality to return polars Dataframes?

what is the expectation if currently some of the estimators are coded to return pandas dataframes. As an example, in regression.mapie, the _predict function turns y_pred into a pandas Dataframe and returns that as an output.

afaik, the boilerplate layer, i.e., BaseProbaRegerssor.predict, will turn it back into a polars.DataFrame if that was passed initially to fit. In the tests, I would check that this is indeed the case.

let me know if/when you would like a review (even if not finished)

Some of the CLI failures look weird - they are just long empty pages.

If i am trying to test an estimator say in test_polars_eager_regressor_in_fit_predict, what is the best approach to do so?