Parallel chains

[leap]

Hello,

I am using MLwiN through runmlwin to run multiprocess models with MCMC estimation. I would like to speed up the estimation process and after considering and reading about several alternatives, I would like to try to run parallel chains. However, I couldn't find any recommendation on how to do parallel chains with MLwiN.

Any material or information would be welcome!

Thank you all in advance.

Léa Pessin

[ChrisCharlton]

Probably the best way to do this is to run your model with different starting values in several instances of Stata and then use the

Code: Select all

mcmcsum, getchains

command to extract the MCMC chains into a datafile. Once these have all run you can merge your datafiles into one and use the

Code: Select all

mcmcsum, variables

command to get the estimates, etc from the chains. We currently don't provide a command to do multichain diagnostics (for example BGRD), however you may be able to use some of the command provided with the -winbugsfromstata- package (http://www2.le.ac.uk/departments/health ... from-stata).

I did write a prefix command some years ago to automate running multiple chains with runmlwin, however this runs the models in series and so wouldn't provide any speedup of the estimation.

[leap]

Thanks for the quick answer.

That's what I have been doing so far but I wasn't sure whether I could use

Code: Select all

mcmc, variables

so that's good to know. Do you have any recommendation on how to choose starting values?

I'll also look into winbugs and, if I find anything useful, I'll report back on the forum in case it comes in handy for anyone else.