BUG: _SearchLight when 'scoring=roc_auc' & y not in (0, 1) #3697
Description
@kaichogami we missed a necessary feature when adding the scoring parameter to the SearchLight in the last PR: currently, we encounter an error if scoring='roc_auc' but y is not (0, 1):
import numpy as np
from mne.decoding.search_light import _SearchLight
from sklearn.linear_model import LogisticRegression
sl = _SearchLight(LogisticRegression(), scoring='roc_auc')
X = np.random.rand(100, 20, 2)
y = np.random.randint(0, 2, 100)
sl.fit(X, y)
sl.score(X, y) # works
y = np.random.randint(2, 4, 100)
sl.fit(X, y)
sl.score(X, y) # crashesI suggest to add a sklearn.preprocessing.LabelEncoder() at fitting and use it before scoring. @agramfort do you think it's the right approach?