hetGP Regression Learner

Heteroscedastic Gaussian process regression. Calls hetGP::mleHetGP() from package hetGP.

Predictions return the posterior mean and the square root of the full predictive variance (sd2 + nugs) as standard error.

noiseControl and settings are represented by explicit learner hyperparameters tagged "noise_control" and "settings", respectively. The wrapper reconstructs the nested lists expected by hetGP::mleHetGP() internally.

Creates a new instance of this learner.

Initial Parameter Values

• return_matrices: Set to FALSE (upstream default is TRUE) to avoid storing inverse covariance matrices in the fitted model. Prediction recomputes them on demand when needed.