Evaluate quality of uncertainty estimates.
Measures correlation between uncertainty and prediction error.
Good uncertainty should be high when the model is wrong.
- Parameters:
uncertainties (Tensor) – Predicted uncertainties (batch_size,)
errors (Tensor) – Binary error indicators (batch_size,) where 1=error, 0=correct
- Return type:
Tuple[float, float]
- Returns:
Tuple of (spearman_correlation, auroc)
