Hi all,
I thought I'd post an example of Metropolis-Coupled MCMC (MCMCMC) that Ian Taylor and I put together (available
here). It's designed to improve mixing for highly multimodal MCMC problems, and can be easily adapted to other applications by adding one line in any given TPL (we demonstrate it using a mixture of bivariate normals). It coordinates ADMB's MCMC utility for unheated and heated sampling chains from within R, and could be parallelized easily.
I post it in case it's helpful to other people, either immediately or as rubric for MCMC changes within the ADMB source code. I'm open to questions, suggestions, or comments.
Cheers,
Jim