probly.conformal_prediction.scores.cqr.torch

PyTorch implementation for CQR scores.

Functions

cqr_score_torch(y_true, y_pred)

PyTorch implementation of CQR nonconformity score.

probly.conformal_prediction.scores.cqr.torch.cqr_score_torch(y_true, y_pred)

PyTorch implementation of CQR nonconformity score.

Computes: s = max(q_lo - y, y - q_hi, 0)

This implementation preserves gradients for backpropagation.

Parameters:

• y_true (Tensor)

• y_pred (Tensor)

Return type:

Tensor