Calibrated Yule model with low ESS after 1 billion generations

[Ana Bedoya]

Hello all,

I am  running a divergence dating analysis in *Beast. I have 43 individuals and 208 concatenated genes. By 1 billion generations, most of my parameters have high ESS values, except for the prior (63), posterior (66), birthRand calibrated yule model (64). I am using a lognormal posterior on a fossil calibration as the crown age for my in-group.

Attached are my results in Tracer. Any ideas on how to speed up the analyses will be appreciated!

Best,

Ana

[Rui Min]

 I also encountered a similar problem. Did you solve it?

[J R]

Hi Ana,

Are you using a birth-death model? If so, that could cause that everything mixes well except for the prior and the posterior. Also, if that is the case, I would say it could be better for you to use a Yule model. I would be happy to hear if implementing this change solves your problem!.

Best,

Juan

[Remco Bouckaert]

HI Ana,

Since the birth rate and ucldMean parameters have low ESS, perhaps increasing the operator weights on these parameters helps a little. Otherwise, just running it longer, or using MC3 (with the CoupledMCMC package https://github.com/nicfel/CoupledMCMC, tutorial: https://taming-the-beast.org/tutorials/CoupledMCMC-Tutorial/) may help with mixing.

Cheers,

Remco

--
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 view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/cbd55f68-8ef2-4607-8a1d-17fdf57d7280o%40googlegroups.com.
<Screen Shot 2020-10-06 at 2.51.32 PM.png>