add .skb.iter_cv_splits()

[jeromedockes]

We have .skb.train_test_split() to do a single train/test split. this adds the equivalent for cross-validation splitters:

import skrub
from sklearn.datasets import make_classification
from sklearn.dummy import DummyClassifier
from sklearn.metrics import accuracy_score
import polars as pl

X_a, y_a = make_classification()
df = pl.DataFrame(X_a).with_columns(y=y_a)

data = skrub.var("data")
X = data.drop("y", strict=False).skb.mark_as_X()
y = data["y"].skb.mark_as_y()
pred = X.skb.apply(DummyClassifier(), y=y)

for split in pred.skb.iter_cv_splits({"data": df}):
    learner = pred.skb.make_learner()
    learner.fit(split["train"])
    prediction = learner.predict(split["test"])
    print("accuracy", accuracy_score(split["y_test"], prediction))

prints

accuracy 0.35
accuracy 0.3
accuracy 0.35
accuracy 0.45
accuracy 0.35

[rcap107]

Looks good to me, thanks a lot @jeromedockes