FIX Array validation for DecisionBoundaryDisplay

[ArturoAmorQ]

Reference Issues/PRs

What does this implement/fix? Explain your changes.

DecisionBoundaryDisplay expects the input data X to be of shape (n_samples, 2), but this is not ensured by an internal validation. Because of this, the following code will output an unintuitive ValueError message:

import numpy as np
import pandas as pd
from sklearn.neighbors import KNeighborsClassifier
from sklearn.inspection import DecisionBoundaryDisplay

rng = np.random.RandomState(0)
X = pd.DataFrame(np.random.randn(5, 4))
y = pd.Series(rng.randint(0, 2, 5))

model = KNeighborsClassifier()
model.fit(X, y)

DecisionBoundaryDisplay.from_estimator(estimator=model, X=X)

Even worse, a further

from sklearn.ensemble import HistGradientBoostingClassifier

model = HistGradientBoostingClassifier()
model.fit(X, y)

DecisionBoundaryDisplay.from_estimator(estimator=model, X=X)

makes a plot!

This PR is a simple fix to solve the issue.

Any other comments?

A meta comment: I am wondering if we could add a parameter column_values (or similar) that accepts dictionaries to fix values for n_features - 2 of the variables, allowing then for surface curves of the decision boundary display.

[glemaitre]

Otherwise LGTM.

[thomasjpfan]

Thank you for the PR @ArturoAmorQ !

[glemaitre]

Actually, I would consider this as a fix and we should also add an entry in the changelog.
We can make it for the 1.2.0 release.

[jeremiedbb]

LGTM. Thanks @ArturoAmorQ

[ArturoAmorQ]

Any idea of the reason for the Windows install fail?

[jeremiedbb]

It's failing everywhere, but we did not find out why yet