This closes #35 and simplifies feature validation. This comment and this commit go into more depth on the changes here, but the bullet points are:
Remove set_temp_output because we no longer need transformers to pass dataframes into transformed estimators since we can get feature names from the transformer.
Remove the warning logic from TransformedKNeighborsMixin._check_feature_names. That is now handled by calling transform_._validate_data. This method now just serves to override the internal check that would incorrectly fail when our estimator is fit with transformed features that mismatch the feature_names_in_ from the transformer.
Update tests to make sure we can fit with a dataframe with no feature names without getting a warning.
Parametrize a few related test cases just to improve error reporting.
This closes #35 and simplifies feature validation. This comment and this commit go into more depth on the changes here, but the bullet points are:
set_temp_outputbecause we no longer need transformers to pass dataframes into transformed estimators since we can get feature names from the transformer.TransformedKNeighborsMixin._check_feature_names. That is now handled by callingtransform_._validate_data. This method now just serves to override the internal check that would incorrectly fail when our estimator is fit with transformed features that mismatch thefeature_names_in_from the transformer.