Hamming loss for multilabel

The problem, from first principles

In multilabel classification one example can carry several tags at once: a photo may be both "beach" and "sunset". Exact-match accuracy counts a row as right only if every tag is right, which is harsh. Hamming loss counts the fraction of individual label slots that are wrong, so partial credit is given for getting most tags right.

Implement hamming_loss(y_true, y_pred). Both inputs are either 1-D label vectors (single-label) or 2-D binary matrices of shape (n_samples, n_labels). Return the fraction of entries where they differ.

Constraints

• The two inputs must have the same shape, otherwise raise ValueError.
• Empty input raises ValueError.
• The result lies in [0, 1].

Hints

Open one at a time. Each gives away a little more than the last.