cqr_r_score

probly.conformal_scores.cqr_r_score = <flextype.singledispatch.Flexdispatch object>

Compute the CQR-r nonconformity score.

The score is the CQR score normalized by the predicted interval width:

s(x, y) = max(q_lo - y, y - q_hi) / max(q_hi - q_lo, ε)

This makes the score adaptive: wider predicted intervals yield smaller normalized scores, rewarding models that express higher uncertainty.

Parameters:
  • y_pred – Predicted lower and upper quantiles, shape (n_samples, 2).

  • y_true – True target values, shape (n_samples,).

Returns:

One-dimensional array of nonconformity scores, shape (n_samples,).

Examples using probly.conformal_scores.cqr_r_score

Quantile Regression Conformal Prediction — sklearn

Quantile Regression Conformal Prediction — sklearn

Quantile Regression Conformal Prediction — PyTorch

Quantile Regression Conformal Prediction — PyTorch