Score probabilistic forecasts with the losses of Murphy's book. Labels are integers.
brier_multiclass(probs, y): probs has shape (N, C). Return with the one-hot label.brier_binary(p, y): for predicted probabilities p of class 1 and 0/1 labels, .brier_skill_score(p, y, p_ref=None): with the binary Brier score. The reference forecasts are p_ref (an array) or, when p_ref is None, the in-sample climatology: the constant for every example. Return when .log_loss(probs, y, eps=1e-15): mean of over the rows of probs (shape (N, C)), after clipping probabilities to .expected_binary_score(p, q, rule): the expected loss of reporting probability for an event whose true probability is . rule is "log" (), "brier" () or "zero_one" (the expected error of predicting if and if ; for use ).Return floats.