Metrics Reference

COREMat evaluates models using the metrics below. Click any name for the full page.

Metric	Key	Description	Compute cost
MAE	mae	Mean Absolute Error — average absolute deviation	Low
RMSE	rmse	Root Mean Squared Error — penalises large errors	Low
MedAE	medae	Median Absolute Error — robust to outliers	Low
R²	r2	Coefficient of Determination — explained variance fraction	Low
R²(log)	r2_log10	R² computed in log₁₀ space	Low
RMSE(log)	rmse_log10	RMSE computed in log₁₀ space	Low
RMSLE	rmsle	RMSE on the log1p scale — relative error, defined at zero	Low
MedAE(log)	medae_log10	MedAE computed in log₁₀ space	Low
EVS	evs	Explained Variance Score — like R² without bias sensitivity	Low
MAPE	mape	Mean Absolute Percentage Error	Low
SMAPE	smape	Symmetric MAPE — symmetric over/under-estimation	Low
MaxErr	me	Maximum Error — worst-case absolute error	Low
OME	ome	Order-of-Magnitude Error — mean log₁₀ distance	Low
OOM Accuracy	oom_acc_x{k}	Fraction within ×k orders of magnitude	Low
Skill Score	—	Weighted composite score relative to a constant baseline	Medium

Common reference example

All per-metric pages use the same running example:

\[ y_{\text{true}} = [5,\ 20,\ 50,\ 100,\ 200] \quad y_{\text{pred}} = [6,\ 18,\ 55,\ 90,\ 220] \quad e_i = y_{\text{true},i} - y_{\text{pred},i} = [-1,\ 2,\ -5,\ 10,\ -20] \]