Regression
Quantile (pinball) loss
Implemented
regression.quantile_lossDefinition
Asymmetric loss used to evaluate quantile predictions.
Formula
Lτ(e) = max(τe, (τ − 1)e), e = y − ŷ
Range: [0, ∞)
Inputs and outputs
- y_true: array of real targets
- y_pred: array of real predictions
- sample_weight: optional
- quantile τ in (0, 1)
Returns: float
Assumptions
No assumptions beyond valid, aligned inputs of the documented types.
Limitations
No metric-specific limitations are documented yet. Interpret the value alongside the task, data, and other metrics.
Python API
import evalsuite as es
es.quantile_loss(y_true_r, y_pred_r, alpha=0.9)References
- Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica, 46(1), 33–50.