GPREG Procedure

The GPREG procedure performs Gaussian process regression (GPR) for very large data sets in SAS Viya.

GPR assumes that the predictions are based on a latent function that is a posterior distribution of a Gaussian process prior and a Gaussian noise likelihood. This latent function is adaptively learned on the basis of the observed data input in the reproducing kernel Hilbert space (RKHS), which is defined by the kernel of the Gaussian process. Thus it provides good regression performance.

This procedure uses stochastic variational inference for Gaussian process models, which can process large data sets fast (Hensman, Fusi, and Lawrence 2013), and where the Gaussian processes can be variationally decomposed to depend on a set of globally relevant inducing variables that factorize the model in the manner necessary to perform variational inference in parallel. This enables you to apply Gaussian process models to data sets that contain millions of data points.