scikit-learn-contrib / category_encoders

A library of sklearn compatible categorical variable encoders
http://contrib.scikit-learn.org/category_encoders/
BSD 3-Clause "New" or "Revised" License
2.39k stars 393 forks source link

Simplify BaseEncoder mapping #403

Open julia-kraus opened 1 year ago

julia-kraus commented 1 year ago

Current State The class variable mapping of the BaseEncoder class is a dictionary with the following structure a dictionary with the following item:

Suggestion Avoid pandas data frames and just use dictionaries instead. These are faster, easier to read and easier to manipulate. Mappings could be chained more easily. Moreover, many of the encoders, they are converted to dict anyways. Use a nested dictionary of the structure {colname: {'cat_a': mapping_val_a, 'cat_b': mapping_val_b, 'cat_c': mapping_val_c ,...)

Actually the whole encoder could just yield the mapping dictionary, because the column names can be retrieved by mapping.keys().

This would really be helpful for better development and readability, but would be work to update and test all subclasses. I think it is unnecessary that users can supply both dictionaries and pd.Series as mappings because they can run pd.Series.to_dict() themselves.

PaulWestenthanner commented 1 year ago

Hi Julia, thanks for your suggestion. I think you're right and we should do it. This would be a major simplification but also a major change in the API. Maybe we can put it in the backlog for version 3.0 together with more simplifications and restructuring that we've been discussing lately