low ESS for likelihood

[Hubert]

Dear all,

I'm puzzled by my results: a BEAST analysis for an empty alignment with 101 terminals, 95 of them assigned attribute "Ingroup" = 1, the other 16 "Ingroup" = 0, HKY model, strict clock, Yule process, standard priors. No additional constraints. Run for a chain length of 1000000000, 4 independent runs from a random starting tree. All parameters have excellent ESS (>7500), except the likelihood and the treeLikelihood, which have an ESS of 5 to 9. The traces are all hairy caterpillars, but likelihood seems not to converge, as in the picture. 

Any hints what's going on?

Hubert Turner

[oryctolagus]

I have the same problem, did you find the solution?

[Brittany Rife]

I am also having the same problem for 1 billion runs for several different data sets. Granted my data sets are a bit large, I didn't expect my TreeLikelihood and Likelihood ESS values to be this small. I am using v1.8, so this should not be an operator issue, correct?

[Guy Baele]

Does 'empty alignment' mean what I think it means? Can you please explain?

Best regards,

Guy

Op zondag 16 maart 2014 15:59:33 UTC-7 schreef Hubert:

[Guy Baele]

Brittany,

You may want to post some more information on your problem, or show some examples of what you're seeing in Tracer (like the first post).

Best regards,

Guy

Op donderdag 7 augustus 2014 17:44:37 UTC-7 schreef Brittany Rife:

[Alexei Drummond]

You need to put a proper prior on the yule.birthrate. a birth rate of 5e99 will only allow trees of miniscule height. You can't sample the prior if it is improper. You need a better prior on the birth rate that keeps it around a priori reasonable values (i.e. the birth rate should be on the order of the reciprocal of an average branch length in substitutions if you are doing an uncalibrated analysis).

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To post to this group, send email to beast...@googlegroups.com.
Visit this group at http://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.