liGP

Performs locally induced approximate GP regression for large computer experiments and spatial datasets following Cole D.A., Christianson, R., Gramacy, R.B. (2021) Statistics and Computing, 31(3), 1-21, . The approximation is based on small local designs combined with a set of inducing points (latent design points) for predictions at particular inputs. Parallelization is supported for generating predictions over an immense out-of-sample testing set. Local optimization of the inducing points design is provided based on variance-based criteria. Inducing point template schemes, including scaling of space-filling designs, are also provided.