Validation Functions

Metrics and diagnostics for surrogate evaluation.

metrics()

from prandtl import metrics
report = metrics(Y_true, Y_pred)
# → {"CL": {"r2": 0.999, "rmse": 0.001, "mae": 0.001, ...}}

Parameter	Type	Description
Y_true	dict[str, ndarray]	Ground truth, keyed by output name
Y_pred	dict[str, ndarray]	Predictions, same keys

Return metric	Description
r2	Coefficient of determination (1 = perfect)
rmse	Root mean square error
mae	Mean absolute error
max_re	Maximum residual error
explained_variance	Explained variance score

cross_validate()

from prandtl import cross_validate
scores = cross_validate(surrogate, X, Y, cv=5)

Parameter	Type	Default	Description
surrogate	Surrogate	—	Trained surrogate
X	ndarray	—	Input data
Y	dict	—	Output data
cv	int	5	Number of folds
verbose	bool	False	Print progress

Returns per-output mae_mean and mae_std across folds.

residual_analysis()

from prandtl import residual_analysis
res = residual_analysis(Y_true, Y_pred)

Returns per-output: shapiro_p (normality), skewness, kurtosis, max_residual_idx.

learning_curve()

from prandtl import learning_curve
curve = learning_curve(surrogate, X, Y, sizes=[20, 50, 100, 200])

Returns train_sizes, train_mae, val_mae for each training size.