BPt.EvalResultsSubset.get_inverse_fis

EvalResultsSubset.get_inverse_fis(fis=None)

Try to inverse transform stored feature importances (either beta weights or automatically calculated feature importances) to their original space.

Warning

If there are any underlying non-recoverable transformations in the pipeline, this method will fail! For example, if a PCA was applied, then a reverse transformation cannot be computed.

This method can be especially helpful when using Loader.

Returns

inverse_fislist of pandas Series

The inverse feature importances will be returned as a list, where each index of the list refers to a fold of the cross-validation, and each element of the list is either a pandas Series or a list of pandas Series (in the case of a categorical problem type where separate feature importances were calculated for each class).

If a Loader was used, the returned Series may contain multi-dimensional arrays instead of scalar values, representing feature importances as transformed back into the original loaded space / shape.