Working on a different problem yesterday I discovered a relatively minor bug in the internally called MCMC functions for fitBayes, the function that implements my new Bayesian method for comparative model-fitting with intraspecific variation with Graham Reynolds (described in our 'in press' article, here). The problem was basically that when MCMC rejected a proposed new set of parameter values, it nonetheless stores the updated likelihood to memory, and returns it at the end of function execution. This is not as serious as it seems because the likelihood was not used in calculations for the next generation of the MCMC, and (especially for a long sample interval, such as the default 100 generations) will usually be numerically trivially different from the correct likelihood (although this doesn't affect our posterior sample, it could affect tests for convergence).
I have posted the new versions of these functions to my phytools page; but I have also built a new nonstatic version of phytools, to which I have also added the new version of ltt described here. You can download the source code for this new version (phytools 0.1-73) here.
No comments:
Post a Comment
Note: due to the very large amount of spam, all comments are now automatically submitted for moderation.