allow passing additional arguments to scorers in dataops

[jeromedockes]

We can now call .skb.with_scoring to define which scorer should be used and pass additional arguments to it (such as sample weights). those can be dataops (ie sample weights are computed dynamically from the inputs)

>>> import skrub
>>> from sklearn.dummy import DummyClassifier

>>> df = skrub.datasets.toy_products()
>>> data = skrub.var("df", df)
>>> X = data[["description", "price"]].skb.mark_as_X(cv=2)
>>> y = data["category"].skb.mark_as_y()
>>> sample_weight = X["price"]

>>> pred = X.skb.apply(DummyClassifier(), y=y).skb.with_scoring(
...     "accuracy", kwargs={"sample_weight": sample_weight}
... )

>>> pred.skb.cross_validate()
   fit_time  score_time  test_score
0  0.002466    0.003201    0.888889
1  0.002033    0.002878    0.647059

[rcap107]

Thanks a lot for this PR @jeromedockes, this will be very useful. I left a few comments for clairification (as usual), and a few small cosmetic changes. I think the main thing that is missing is an entry in the API reference. We might also want to update one of the examples in the gallery with the new method at some point.

[jeromedockes]

thanks for the review @rcap107 , I checked the links and they should be good now

[rcap107]

Looks good to me! Thanks a lot @jeromedockes

[jeromedockes]

thanks for the reviews :)