Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
NCTM or The Math Forum.



Re: rjmcmc
Posted:
Nov 21, 2012 5:21 AM


"majeti dinesh" wrote in message news:fd072664a7954aff8a005ee56044dee8@googlegroups.com...
Hi, Does anyone have an idea about the implementation issues with rjmcmc method. I am facing some issues while implementing it in c++. The probability values are beoming too small (e500). Hence resulting in zero output...

It would help if you said what you think the "rjmcmc method" is.
Probabilities should sum to one, so possibilities are
(i) you are forgetting to rescale so that the probabilities do sum to one
(ii) you are working with an extremely large number of possibilities, in which case you might be able to combine some together
(iii) you could work with the logarithms of the probabilties instead.
Of course, everything depends on what the "rjmcmc method" is.



