Multidimension bridge sampling (CoRe in CiRM [5])

[This article was first published on Xi'an's Og » R, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Since Bayes factor approximation is one of my areas of interest, I was intrigued by Xiao-Li Meng’s comments during my poster in Benidorm that I was using the “wrong” bridge sampling estimator when trying to bridge two models of different dimensions, based on the completion (for theta_2=(mu,sigma^2) and mu=theta_1 missing from the first model)

B^pi_{12}(x)= dfrac{displaystyle{intpi_1^*(mu|sigma^2){tildepi}_1(sigma^2|x) alpha(theta_2) {pi}_2(theta_2|x)hbox{d}theta_2}}{ displaystyle{int{tildepi}_2(theta_2|x)alpha(theta_2) pi_1(sigma^2|x)hbox{d}sigma^2 } pi_1^*(mu|sigma^2) hbox{d}mu },.

When revising the normal chapter of Bayesian Core,  here in CiRM, I thus went back to Xiao-Li’s papers on the topic to try to fathom what the “true” bridge sampling was in that case. In Meng and Schilling (2002, JASA), I found the following indication, “when estimating the ratio of normalizing constants with different dimensions, a good strategy is to bridge each density with a good approximation of itself and then apply bridge sampling to estimate each normalizing constant separately. This is typically more effective than to artificially bridge the two original densities by augmenting the dimension of the lower one”. I was unsure of the technique this (somehow vague) indication pointed at until I understood that it meant  introducing one artificial posterior distribution for each of the parameter spaces and processing each marginal likelihood as an integral ratio in itself. For instance, if eta_1(theta_1) is an arbitrary normalised density on theta_1, and alpha is an arbitrary function, we have the bridge sampling identity on m_1(x):

inttilde{pi}_1(theta_1|x) ,text{d}theta_1 = dfrac{displaystyle{int tilde{pi}_1(theta_1|x) alpha(theta_1) {eta}_1(theta_1),text{d}theta_1}}{displaystyle{inteta_1(theta_1) alpha(theta_1) pi_1(theta_1|x) ,text{d}theta_1}}

Therefore, the optimal choice of alpha leads to the approximation

widehat m_1(x) = dfrac{displaystyle{sum_{i=1}^N {tildepi}_1(theta^eta_{1i}|x)big/left{{m_1(x) tildepi}_1(theta^eta_{1i}|x) + eta(theta^eta_{1i})right}}}{displaystyle{ sum_{i=1}^{N} eta(theta_{1i}) big/ left{{m_1(x) tildepi}_1(theta_{1i}|x) + eta(theta_{1i})right}}}

when theta_{1i}simpi_1(theta_1|x) and theta^eta_{1i}simeta(theta_1). More exactly, this approximation is replaced with an iterative version since it depends on the unknown m_1(x). The choice of the density eta is obviously fundamental and it should be close to the true posterior pi_1(theta_1|x) to guarantee good convergence approximation. Using a normal approximation to the posterior distribution of theta or a non-parametric approximation based on a sample from pi_1(theta_1|mathbf{x}), or yet again an average of MCMC proposals are reasonable choices.

The boxplot above compares this solution of Meng and Schilling (2002, JASA), called double (because two pseudo-posteriors eta_1(theta_1) and eta_2(theta_2) have to be introduced), with Chen, Shao and Ibragim (2001) solution based on a single completion pi_1^* (using a normal centred at the estimate of the missing parameter, and with variance the estimate from the simulation), when testing whether or not the mean of a normal model with unknown variance is zero. The variabilities are quite comparable in this admittedly overly simple case. Overall, the performances of both extensions are obviously highly dependent on the choice of the completion factors, eta_1 and eta_2 on the one hand and pi_1^* on the other hand, . The performances of the first solution, which bridges both models via pi_1^*, are bound to deteriorate as the dimension gap between those models increases. The impact of the dimension of the models is less keenly felt for the other solution, as the approximation remains local.


Filed under: Books, R, Statistics, University life Tagged: Bayesian Core, Benidorm, bridge sampling, CIRM, poster, Valencia 9

To leave a comment for the author, please follow the link and comment on their blog: Xi'an's Og » R.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)