search

beast-dev / beast-mcmc

Bayesian Evolutionary Analysis Sampling Trees
http://beast.community
GNU Lesser General Public License v2.1
192 stars 73 forks source link

log both allNus and allMus? #1022

Closed GuyBaele closed 6 years ago

GuyBaele commented 6 years ago

Using the new parameterization, we're operating on the 'allNus' but still logging the 'allMus' which is kind of strange to me. Additionally, this leaves me puzzled as to how generalized stepping-stone sampling should work? Ideally, I'd like to construct working priors using the 'allNus', so that would mean logging them. Is it out of the question to log both?

maxbiostat commented 6 years ago

The mu's are of direct scientific interest, since they are the relative weights of the evolutionary rate. The nu's are merely a computational trick that supposedly makes sampling more efficient.

rambaut commented 6 years ago

No reason why we couldn't log _nu_s as well as _mu_s.

maxbiostat commented 6 years ago

Yeah, but logging the nu's serves no purpose to the end user, unless they're interested in the detecting mixing problems. My understanding is that the nus are just a computational trick; they're hard to interpret. Maybe best to avoid confusion and just log the mus.

rambaut commented 6 years ago

Didn't mean to close it.

GuyBaele commented 6 years ago

Logging both for the moment.