Automated Factor Slice Sampling

• Activity Number: 302
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Matthew Tibbits*+ and Chris Groendyke and Murali Haran
• Companies: Penn State and Penn State and Penn State
• Keywords: Slice Sampling ; Markov chain Monte Carlo ; Automated Tuning Procedure ; Eigenvector Decomposition

Abstract

Markov chain Monte Carlo (MCMC) algorithms are general approaches for sampling from arbitrary distributions. However, devising and tuning MCMC algorithms for each new distribution can be challenging and time consuming. It is particularly difficult when there is strong dependence among the variables in a multivariate distribution. We describe a two-pronged approach for automating MCMC algorithms: (i) we propose the "factor slice sampler", a generalization of the univariate slice sampler where we treat the selection of a coordinate basis (factors) as an additional tuning parameter, and (ii) we develop an approach for automatically finding tuning parameters in order to construct an efficient factor slice sampler. Our tuning approach also applies to standard univariate slice samplers. We demonstrate the application of our automated MCMC algorithm in the context of several examples.