How to automatically select the nugget parameter in Gaussian process regression (GPR)?

Question

In sklearn's GPR function, how can I automatically determine the nugget parameter? It does not seem to be implemented within the package. Any insights are welcome.

Answer 1

The nugget, as it is called in geostatistics, corresponds to the White-noise kernel in sklearn. If your kernel includes a "nugget" term, its noise-level parameter will be automatically calibrated (via mle) by the fit() method of a GaussianProcessRegressor estimator.

For details, see the example "GPR with noise-level estimation" in the sklearn docs.

Answer 2

You can use any general hyperparameter selection such as Grid Search.

http://scikit-learn.org/stable/modules/grid_search.html