Ensure predict* and transform preserve indexes on dataframes

KhiopsML / khiops-python

The Python library of the Khiops AutoML suite

https://khiops.org

BSD 3-Clause Clear License

9 stars 1 forks source link

Ensure predict* and transform preserve indexes on dataframes #104

Open folmos-at-orange opened 1 year ago

folmos-at-orange commented 1 year ago

Description

Currently the indexes is lost because we read from an internal file. Sklearn's new set_output API seems to respect the index of the input (see examples here). Additionally it simplifies the UX as the user doesn't need to reindex its input to join input and output dataframes.

Questions/Ideas

In the predict* and transform methods we have available the index of the input. At the end, after reading the output file from Khiops we just reindex using that one.

folmos-at-orange commented 2 weeks ago

Putting this in stand by because now we return numpy.arrays. Once we support the set_output sklearn API we can tackle this (see #43)

popescu-v commented 2 weeks ago

Putting this in stand by because now we return numpy.arrays. Once we support the set_output sklearn API we can tackle this (see #43)

Yes, this seems like a longer-term feature.

folmos-at-orange commented 1 week ago

As mentioned in #43 this will concern only KhiopsCoclustering and KhiopsEncoder.