GADGET: Gaussian Process Approximations for Designing Experiment

In GADGET: Gaussian Process Approximations for Designing Experiments

Description

The GADGET package computes near-optimal Bayesian experimental designs using Gaussian process optimization. At its core is the ability to calculate static designs that maximize a design criterion that may be either deterministic or stochastic. In particular, stochastic design criteria could be a Monte Carlo estimator of an expected utility based on MCMC posterior draws. GADGET utilizes the algorithm proposed by Weaver et al. (2016) <doi:10.1214/15-BA945> and performs Gaussian process validation using the statistics introduced by Bastos and O’Hagan (2009) <doi:10.1198/TECH.2009.08019>. The pbapply package is integrated to parallelize the evaluation of the user's design criterion. Additionally, GADGET has wrapped the optimization into a sequential routine to perform sequential computer experiments that automatically call simulator code that is available in R.

Author(s)

Isaac Michaud, Brian Weaver, and Brian Williams

References

Weaver, B. P., Williams, B. J., Anderson-Cook, C. M., Higdon, D. M. (2016). Computational enhancements to Bayesian design of experiments using Gaussian processes. Bayesian Analysis, 11(1), 191–213, <doi:10.1214/15-BA945>.

GADGET documentation built on Jan. 25, 2020, 1:06 a.m.