Regression

Available in v0.1.0

Usage

Pythonv0.1.0

import evalsuite as es

es.mae(y_true, y_pred)
es.rmse(y_true, y_pred, sample_weight=w)
es.mape(y_true, y_pred)   # raises MetricInputError if any target is zero

es.evaluate(y_true, y_pred)   # MAE, MSE, RMSE, R² and more in one call

Single and multi-output targets are supported (multioutput="uniform_average", "raw_values" or weights).

Domain restrictions

Some metrics are only defined on part of the real line. Inputs are checked before computation:

MAPE requires every target to be non-zero.
MSLE and RMSLE require targets and predictions greater than −1.
R² is undefined for constant targets.

Invalid inputs raise MetricInputError instead of producing infinities or misleading values.

Metrics

Regression metrics
Metric	Description	Status	API
Mean absolute error	Average absolute difference between predictions and targets, in target units.	Implemented	es.mae
Mean squared error	Average squared difference between predictions and targets.	Implemented	es.mse
Root mean squared error	Square root of MSE, expressed in target units.	Implemented	es.rmse
Coefficient of determination (R²)	Proportion of variance in the target explained by the predictions, relative to predicting the mean.	Implemented	es.r2
Adjusted R²	R² penalised for the number of predictors in the model.	Implemented	es.adjusted_r2
Mean absolute percentage error	Average absolute error relative to the magnitude of the target.	Implemented	es.mape
Symmetric MAPE	Percentage error scaled by the mean magnitude of target and prediction.	Implemented	es.smape
Root mean squared log error	RMSE computed on log(1 + value), emphasising relative rather than absolute error.	Implemented	es.rmsle
Median absolute error	Median of absolute errors; robust to outliers.	Implemented	es.median_absolute_error
Huber loss	Quadratic for small errors and linear for large errors, controlled by δ.	Implemented	es.huber_loss
Quantile (pinball) loss	Asymmetric loss used to evaluate quantile predictions.	Implemented	es.quantile_loss

Mean absolute errorImplemented
Average absolute difference between predictions and targets, in target units.
es.mae
Mean squared errorImplemented
Average squared difference between predictions and targets.
es.mse
Root mean squared errorImplemented
Square root of MSE, expressed in target units.
es.rmse
Coefficient of determination (R²)Implemented
Proportion of variance in the target explained by the predictions, relative to predicting the mean.
es.r2
Adjusted R²Implemented
R² penalised for the number of predictors in the model.
es.adjusted_r2
Mean absolute percentage errorImplemented
Average absolute error relative to the magnitude of the target.
es.mape
Symmetric MAPEImplemented
Percentage error scaled by the mean magnitude of target and prediction.
es.smape
Root mean squared log errorImplemented
RMSE computed on log(1 + value), emphasising relative rather than absolute error.
es.rmsle
Median absolute errorImplemented
Median of absolute errors; robust to outliers.
es.median_absolute_error
Huber lossImplemented
Quadratic for small errors and linear for large errors, controlled by δ.
es.huber_loss
Quantile (pinball) lossImplemented
Asymmetric loss used to evaluate quantile predictions.
es.quantile_loss