Software Code Fast Computing Gaussian Process Regression

Gaussian Process (GP) regression is a popular Bayesian nonparametric approach for non-liner regression analysis.  It has many useful applications in remote sensing, spatial data analysis, and simulation meta-modeling.  However, its computation is prohibitively expensive when the amount of data is very large.  This software code implements an inexpensive approximate computation algorithm to achieve a Gaussian Process regression solution quickly and accurately.  The name of the approximation algorithm is the Patchwork Kriging.  The inventor developed it in 2019. 

The Patchwork Kriging involves partitioning the regression input data into multiple local regions with a different local Gaussian Process models that are fitted in each specific region.  Unlike existing local Gaussian Process models, this application introduced a technique which can patch together the local Gaussian Process models nearly seamlessly to ensure that the local Gaussian Process models for two neighboring regions produce nearly the same response prediction and prediction error variance on the boundary between the two regions.  This largely mitigates well known discontinuity problems that tend to degrade the prediction accuracy of existing locally partitioned Gaussian Process methods over regional boundaries. 

Advantages

• Accuracy in estimations
• Differentiation between neighboring area and their calculations