`scikit-learn 1.7.0` compatibility

[fkiraly]

scikit-learn 1.7.0 has been released recently, and extension contracts have changed, in particular:

• the tag system
• the validate_data logic

Is this being worked on, or would this be a useful contribution?

[solegalli]

Thank you @fkiraly

Yes, every time sklearn makes a new release it is a real headache. We need to update the library to reflect the new release changes.

In the last sklearn release, we tried to make feature engine compatible with newer and older versions, because lack of compatibility with older versions is also a headache for users. Not sure it is possible to maintain that for this release as well. I need to look into it. But if you have a solution, please go for it :)

[fkiraly]

But if you have a solution, please go for it :)

Yes, we had the same problem in sktime, skpro, and hyperactive.

My most recent thinking is, one probably needs an intermediary mixin that is version stable - this is what we have been using in sktime:
https://github.com/sktime/sktime/blob/main/sktime/utils/sklearn/_version_bridge.py

This makes explicit version checks internally, and delegates to the correct pattern.

[solegalli]

Ok, looking further into this issue, we did make the library compatible with sklearn 1.7.0 in PRs #833 to #836

Feature-engine does not use sklearn validate_data function, we have our own validation module.

Is this PR related to the failed tests that we see in PR #862 ?

[fkiraly]

Feature-engine does not use sklearn validate_data function, we have our own validation module.

Oh, nice! I have a few questions:

• how do you pass the sklearn tests then?
• is this something we could expose as a general utility for packages extending sklearn? I am having a hard time at every sklearn minor update too, so if you have already figured out how to manage the boilerplate, it could help in multiple packages!

[fkiraly]

Is this PR related to the failed tests that we see in PR #862 ?

Sorry for the confusion: which PR are you referring to?

[solegalli]

Feature-engine does not use sklearn validate_data function, we have our own validation module.

Oh, nice! I have a few questions:

* how do you pass the `sklearn` tests then?

* is this something we could expose as a general utility for packages extending `sklearn`? I am having a hard time at every `sklearn` minor update too, so if you have already figured out how to manage the boilerplate, it could help in multiple packages!

Unfort, feature-engine has a lot of boilerplate thanks to sklearn constant API changes. I have boilerplate in the transformers and in the tests. At some point I will start supporting only the newer versions and remove the boilerplate. That was my idea.