A new result processing module has been added: it allows to perform SiR (sampling-importance-resampling) draws from an approximate posterior distribution of model parameters.

Either asymptotic normal approximation or latin hypercube grid can serve as initial distributions for resampling; effective size of the sample is reported.

Any subset of model parameters can be sampled; the results are written to a tab-separated file (and opened in a chart window on GUI builds).

For models with rate variation, conditional probabilities for all sites can be written out for all samples.

Dear Sergei,

One more thing: how about a nonparametric MLE of the likelihood surface? The normal sampler is sometimes a bit too stringent, and the grid sampler can give very low effective sample sizes.

Best
Simon