geopfa.extrapolation.assess_gp_model_fit

assess_gp_model_fit(model, Y_true, Y_pred, Y_pred_std, constraints)[source]

Compute regression performance metrics and model diagnostics for a trained Gaussian Process model.

Parameters:

  • model (GPy.models.GPRegression) – Trained GP regression model.

  • Y_true (numpy.ndarray) – True observed target values. Flattening is handled internally.

  • Y_pred (numpy.ndarray) – Predicted mean values from the GP model (same shape as Y_true).

  • Y_pred_std (numpy.ndarray) – Predicted standard deviations for the GP predictions.

  • constraints (dict) – Constraint information returned during GP construction, used to check whether any hyperparameters lie on their bounds.

Returns:

dict – A dictionary with the following entries:

  • RMSEfloat

    Root Mean Square Error.

  • R2float

    Coefficient of determination.

  • MAEfloat

    Mean Absolute Error.

  • Coverage_95float

    Fraction of observations lying inside ±2 sigma predictive intervals.

  • LogLikelihoodfloat

    GP model log marginal likelihood.

  • AICfloat

    Akaike Information Criterion 2*k - 2*logL.

  • BICfloat

    Bayesian Information Criterion k*ln(n) - 2*logL.

  • Params_at_boundslist of tuple

    Parameters effectively pinned at their allowable bounds.

Notes

  • No transformations or standardization adjustments are performed here. Inputs should already be in the desired scale.

  • The ±2 sigma coverage statistic provides a simple approximate 95% interval assessment for GP predictive uncertainty.