uacqr_score¶
- probly.conformal_scores.uacqr.uacqr_score = <flextype.singledispatch.Flexdispatch object>¶
Compute the UACQR nonconformity score.
Normalizes the CQR score by the standard deviation of the ensemble’s predicted quantiles across estimations:
std_lo = std(intervals[:, :, 0], axis=0) std_hi = std(intervals[:, :, 1], axis=0) q_lo = mean(intervals[:, :, 0], axis=0) q_hi = mean(intervals[:, :, 1], axis=0) s(x, y) = max((q_lo - y) / std_lo, (y - q_hi) / std_hi)
- Parameters:
y_pred – Ensemble interval predictions, shape
(n_estimations, n_samples, 2), whereintervals[..., 0]are lower quantiles andintervals[..., 1]are upper quantiles.y_true – True target values, shape
(n_samples,).
- Returns:
One-dimensional array of nonconformity scores, shape
(n_samples,).