Fail explicitly on multilabel/multiouput targets

[chkoar]

I believe that we should be more explicit about the multioutput/multilabel format. The first error is helpful from a developers' point of view. The second error makes no sense to a user. So, since we know the target type of the provided dataset we should fail with an explicit message.

from imblearn.under_sampling import RandomUnderSampler
from imblearn.ensemble import BalancedBaggingClassifier, BalancedRandomForestClassifier
from sklearn.datasets import make_multilabel_classification

X, y = make_multilabel_classification()

ros = RandomUnderSampler()
ros.fit_sample(X,y)

ValueError: When 'y' corresponds to 'multilabel-indicator', 'y' should encode th
e multiclass (a single 1 by row).

bagging = BalancedBaggingClassifier()
bagging.fit(X,y)

ValueError: bad input shape (100, 5)

[glemaitre]

It is raise by scikit learn. Nor sure we will pass the common test if we reraise something different