search

beast-dev / beast-mcmc

Bayesian Evolutionary Analysis Sampling Trees
http://beast.community
GNU Lesser General Public License v2.1
181 stars 69 forks source link

Need scaled logit for generalized stepping-stone sampling? #953

Open GuyBaele opened 6 years ago

GuyBaele commented 6 years ago

Currently, log transformation is used for positive parameters and logit for probabilities (like the base frequencies being estimated). The relative rate parameters are quite similar to the base frequencies, but with an upper limit of a fixed value instead of 1. Right now the relative rate parameters received a log transformed reference prior, but might be better to have a scaled logit reference prior.

rambaut commented 6 years ago

Why do the relative rates have a fixed upper limit? Their sum is fixed but no upper limit.