LE3 .A278 2012

2012

Ranjan, Pritam
Chipman, Hugh

Acadia University

Bachelor of Science

Honours

Mathematics and Statistics

Mathematics & Statistics

Gaussian process (GP) models are commonly used for emulating computationally expensive computer simulators. Fitting a GP model requires the computation of the inverse and determinant of the spatial correlation matrix, R. The near-singularity of R can cause instability in the computation of its determinant and inverse. Ranjan et al. (2011) proposed a robust approach to overcome this problem. They used a robust but somewhat slow genetic algorithm based approach for maximizing the likelihood. In this thesis we investigate several parameterizations of the likelihood for ease of optimization. We then develop a new algorithm to nd the global optimum of the likelihood surface. Moreover, we developed an R package called GPfit that imple- ments the proposed methodology. Finally, we use several test functions to compare the performance of our approach with a popular GP tting package in R.

The author retains copyright in this thesis. Any substantial copying or any other actions that exceed fair dealing or other exceptions in the Copyright Act require the permission of the author.

https://scholar.acadiau.ca/islandora/object/theses:924