sklearn .fit method incorrectly applies arg max

[simplymathematics]

Describe the bug
The ScikitlearnClassifier.fit method incorrectly applies argmax when y_labels are already a single column.

The method applies np.argmax without checking the shape.

To Reproduce
Steps to reproduce the behavior:
art_model = SciKitlearnClassifier().fit(X, y=[0,0,0,0,1,1,1,1,1])

Expected behavior
Only apply np.argmax if ndim > 1.

Screenshots
If applicable, add screenshots to help explain your problem.

System information (please complete the following information):
https://github.com/Trusted-AI/adversarial-robustness-toolbox/blob/main/art/estimators/classification/scikitlearn.py

In the current main branch.

[wasibabi]

Hi! I’d be interested in working on this issue if it’s still available.

From my initial investigation, it looks like ScikitlearnClassifier.fit() unconditionally applies np.argmax(y_preprocessed, axis=1), which works for one-hot encoded labels but fails when y is already provided as a 1D vector of class indices (shape == (n_samples,)). Since ART documentation appears to support both formats, I believe a shape-aware conversion may be appropriate here.

I’ll spend some time reproducing the issue locally and, assuming there aren’t any concerns with the proposed approach, I’d be happy to submit a fix along with regression tests.