[R] An important question about running MCMC

[Barry Rowlingson]

On Thu, Dec 13, 2012 at 3:22 AM, Chenyi Pan <cp2xd at virginia.edu> wrote:
> Dear officer
> I have a question concerning running R when I am doing my research. Can you
> help me to figure that out?
>
> I am now running a MCMC iteration in the R program. But it is always
> stucked in some loop. This cause big problems for my research. So I want to
> know whether we can skip the current dataset and move to next simulated
> data when the iteration is stucked? Alternatively, can the MCMC chain skip
> the current iteration when it is stucked and automatically to start another
> chain with different starting values.
>
> I am looking forward to your reply.

 What do you mean by 'stucked in some loop'? Is it so stuck it isn't
generating proposals, because of a bug in your code? Or do you mean it
is generating proposals but never accepting them? In which case maybe
you need to look at your proposal distribution and make your proposal
scheme adaptive. Or is it looping around a set of accepted values? Or
what?

 Remember, for true convergence of an MCMC you have to loop an
infinite number of times - have you tried that?

Barry