[BUG] hinge_loss multi

>>> sklearn.metrics.hinge_loss(y_true=[2,1,0,1,0,1,1], pred_decision=[0,1,2,1,0,2,1])

  2378         y_true = le.transform(y_true)
   2379         mask = np.ones_like(pred_decision, dtype=bool)
-> 2380         mask[np.arange(y_true.shape[0]), y_true] = False
   2381         margin = pred_decision[~mask]
   2382         margin -= np.max(pred_decision[mask].reshape(y_true.shape[0], -1),

IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed

[PierreAttard]

Ok, so I take and raise an Error if the shape of pred_decision is not consistent with y_true (or labels) like said @thomasjpfan.

[NicolasHug]

No, the hinge loss requires the real-valued decision function, not the predicted classes

    In the multiclass case:

    >>> import numpy as np
    >>> X = np.array([[0], [1], [2], [3]])
    >>> Y = np.array([0, 1, 2, 3])
    >>> labels = np.array([0, 1, 2, 3])
    >>> est = svm.LinearSVC()
    >>> est.fit(X, Y)
    LinearSVC()
    >>> pred_decision = est.decision_function([[-1], [2], [3]])
    >>> y_true = [0, 2, 3]
    >>> hinge_loss(y_true, pred_decision, labels=labels)
    0.56...

where pred_decision is

In [3]: pred_decision
Out[3]:
array([[ 1.27271904,  0.03418064, -0.68377391, -1.40170867],
       [-1.45454381, -0.58121055, -0.37602669, -0.17093709],
       [-2.36363142, -0.78634095, -0.27344428,  0.23932011]]