Inspect any scikit-learn fitted Pipeline to transform a feature metadata pandas DataFrame through the Pipeline and add model interpretation.

transform_feature_meta.py

def transform_feature_meta(pipe, feature_meta):
    transformed_feature_meta = None
    for estimator in pipe:
        if isinstance(estimator, ColumnTransformer):
            for _, trf_pipe, trf_columns in estimator.transformers_:
                if isinstance(trf_pipe, str) and trf_pipe == 'drop':
                    trf_feature_meta = feature_meta.iloc[
                        ~feature_meta.index.isin(trf_columns)]
                elif ((isinstance(trf_columns, slice)
                       and (isinstance(trf_columns.start, str)
                            or isinstance(trf_columns.stop, str)))
                      or isinstance(trf_columns[0], str)):
                    trf_feature_meta = feature_meta.loc[trf_columns]
                else:
                    trf_feature_meta = feature_meta.iloc[trf_columns]
                if isinstance(trf_pipe, BaseEstimator):
                    for trf_estimator in trf_pipe:
                        if hasattr(trf_estimator, 'get_support'):
                            trf_feature_meta = trf_feature_meta.loc[
                                trf_estimator.get_support()]
                        elif hasattr(trf_estimator, 'get_feature_names'):
                            trf_new_feature_names = (
                                trf_estimator.get_feature_names(
                                    input_features=(trf_feature_meta.index
                                                    .values)).astype(str))
                            trf_feature_meta = pd.DataFrame(
                                np.repeat(trf_feature_meta.values, [
                                    np.sum(np.char.startswith(
                                        trf_new_feature_names,
                                        '{}_'.format(feature_name)))
                                    for feature_name in trf_feature_meta.index
                                ], axis=0), columns=trf_feature_meta.columns,
                                index=trf_new_feature_names)
                if transformed_feature_meta is None:
                    transformed_feature_meta = trf_feature_meta
                else:
                    transformed_feature_meta = pd.concat(
                        [transformed_feature_meta, trf_feature_meta], axis=0)
        else:
            if transformed_feature_meta is None:
                transformed_feature_meta = feature_meta
            if hasattr(estimator, 'get_support'):
                transformed_feature_meta = (
                    transformed_feature_meta.loc[estimator.get_support()])
            elif hasattr(estimator, 'get_feature_names'):
                new_feature_names = estimator.get_feature_names(
                    input_features=transformed_feature_meta.index.values
                ).astype(str)
                transformed_feature_meta = pd.DataFrame(
                    np.repeat(transformed_feature_meta.values, [
                        np.sum(np.char.startswith(
                            new_feature_names, '{}_'.format(feature_name)))
                        for feature_name in transformed_feature_meta.index
                    ], axis=0), columns=transformed_feature_meta.columns,
                    index=new_feature_names)
    final_estimator = pipe[-1]
    feature_weights = explain_weights_df(
        final_estimator, feature_names=transformed_feature_meta.index.values)
    if feature_weights is None and hasattr(final_estimator, 'estimator_'):
        feature_weights = explain_weights_df(
            final_estimator.estimator_,
            feature_names=transformed_feature_meta.index.values)
    if feature_weights is not None:
        feature_weights.set_index('feature', inplace=True,
                                  verify_integrity=True)
        feature_weights.columns = map(str.title, feature_weights.columns)
        transformed_feature_meta = transformed_feature_meta.join(
            feature_weights, how='inner')
    transformed_feature_meta.index.rename('Feature', inplace=True)
    return transformed_feature_meta

If you do not want to use ELI5, replace https://gist.github.com/hermidalc/0ee8b3846b0ffd368dd6b7f3b310adcc#file-transform_feature_meta-py-L57-L68 with:

feature_weights = None
if hasattr(final_estimator, 'coef_'):
    feature_weights = final_estimator.coef_
elif hasattr(final_estimator, 'feature_importances_'):
    feature_weights = final_estimator.feature_importances_
elif hasattr(final_estimator, 'estimator_'):
    if hasattr(final_estimator.estimator_, 'coef_'):
        feature_weights = final_estimator.estimator_.coef_
    elif hasattr(final_estimator.estimator_, 'feature_importances_'):
        feature_weights = final_estimator.estimator_.feature_importances_
if feature_weights is not None:
    final_feature_meta['Weight'] = feature_weights