[Feature Request] Flexibility in Scikit-Learn version

[jolespin]

Is your feature request related to a problem? Please describe.
Dependency on new-ish Scikit-Learn version makes general usage difficult

Describe the solution you'd like
Find the minimum version of Scikit-Learn that won't break Feature Engine then use that as the baseline.

Describe alternatives you've considered
Copy and pasting classes from Feature Engine into packages that are only compatible with lower Scikit-Learn versions

Additional context
Taking a look at Feature Engine version 1.8.0 and 1.7.0 which requires Scikit-Learn 1.4.0 but Feature Engine version 1.6.2 requires any Scikit-Learn above 1.0.0.

[solegalli]

Hey @jolespin thanks for the request.

If I understand correctly, you'd like to use Fetaure-engine 1.8 for example, with sklearn 1.0.0. Is this correct?

[jolespin]

If it doesn't break feature engine, I think this could be useful for increased compatibility. If feature engine 1.8.0 relies on features in higher sklearn versions then we can close this.

[solegalli]

@ChristopherGS would you be able to modify the tox functionality so that python 11 for example tests twice, once with a version of sklearn = 1.0.0 and then again with sklearn=1.4.0?

It doesn't matter if the tests fail, I think they will for 1.0.0, but then I can fix them. It would be great if you could do that :)

[CGSLivv]

@ChristopherGS would you be able to modify the tox functionality so that python 11 for example tests twice, once with a version of sklearn = 1.0.0 and then again with sklearn=1.4.0?

It doesn't matter if the tests fail, I think they will for 1.0.0, but then I can fix them. It would be great if you could do that :)

Here we go: https://github.com/feature-engine/feature_engine/tree/maintenance/tox-testing-for-sklearn-versions

If you make me a collaborator I can open a PR directly

[solegalli]

implemented in #837

Later versions of feature-engine support scikit-learn 1.6.0 onwards

Older versions of feature-engine support older versions of sklearn.

Sklearn changes are too big for us to cope with, so this is our best for now.