scikit-learn pipeline strategy with both numeric and categorical columns

Question

We are implementing an sklearn pipeline as follows (pseudo code):

feature_num = numeric_columns  
feature_cat = categorical_columns  

num_pipeline = Pipeline([('feature_select', DataFrameSelector(feature_num))
                        ,('feature_num_prep', custom_num_prep)
                       ])  
cat_pipeline = Pipeline([('feature_select', DataFrameSelector(feature_cat))
                        ,('feature_cat_prep', custom_cat_prep)
                       ])  
feature_pipeline = FeatureUnion(transformer_list=[
                               ('num_pipeline', num_pipeline)
                               ,('cat_pipeline', cat_pipeline)
])  

Since we have in the custom_num_prep transformer the possibility to remove outliers - when we get to the feature_pipeline there is an error:

ValueError: [blocks[0,:] has incompatible row dimensions....

The reason is obvious since the cat_pipeline has observations that the num_pipeline has removed.
The question is:
1. is there a pythonic way to sync the two dataframes within the pipelines?
2. if not what strategy is there for such situations?

Answer 1

One option is scikit-learn's ColumnTransformer applied to mixed types. ColumnTransformer is designed for the purpose of applying different preprocessing and feature extraction pipelines to different subsets of the features.