Current browse context:
stat.CO
Change to browse by:
References & Citations
Statistics > Computation
Title: Bayesian model comparison with un-normalised likelihoods
(Submitted on 1 Apr 2015 (v1), last revised 20 Jan 2016 (this version, v3))
Abstract: Models for which the likelihood function can be evaluated only up to a parameter-dependent unknown normalising constant, such as Markov random field models, are used widely in computer science, statistical physics, spatial statistics, and network analysis. However, Bayesian analysis of these models using standard Monte Carlo methods is not possible due to the intractability of their likelihood functions. Several methods that permit exact, or close to exact, simulation from the posterior distribution have recently been developed. However, estimating the evidence and Bayes' factors (BFs) for these models remains challenging in general. This paper describes new random weight importance sampling and sequential Monte Carlo methods for estimating BFs that use simulation to circumvent the evaluation of the intractable likelihood, and compares them to existing methods. In some cases we observe an advantage in the use of biased weight estimates. An initial investigation into the theoretical and empirical properties of this class of methods is presented. Some support for the use of biased estimates is presented, but we advocate caution in the use of such estimates.
Submission history
From: Richard Everitt [view email][v1] Wed, 1 Apr 2015 17:10:25 GMT (325kb,D)
[v2] Wed, 6 Jan 2016 15:42:08 GMT (318kb,D)
[v3] Wed, 20 Jan 2016 23:33:41 GMT (318kb,D)
Link back to: arXiv, form interface, contact.