Low ESS values in posterior, prior and clock rate

[uxia saraiba esperon moldes]

Dear Beast Users, 

I am trying to create an Extended Bayesian Skyline Plot from multiple loci data. The number of samples isnt great so I am trying to run my data for millions of chains which ends up with a super big log file. The problem is that I still get ESS low values around 160 in posterior, prior and meanRate. When setting the prior in Beauti I am giving a fixed value on the clock rate prior and a uniform distribution for the Demographic PopulationMean. I also tried to use the logCombiner to improve the ESS value but the resultant file is just too big!! I usually log every 1000 with a Length of chain of 100000000. Is anyone experiencing the same problem? Any suggestions? Thanks a lot!! 

[Alexei Drummond]

Log every 10,000 instead. I believe logcombiner can thin your log files for you so you don’t need to redo the analysis. It will have no impact apart from reducing the file sizes by a factor of 10. You only need between 1000 and 10,000 logged values, so if you increase the chainlength for your analysis it usually makes sense to decrease the logging frequency.

Cheers

Alexei

On 5/05/2017, at 10:28 am, uxia saraiba esperon moldes <uxias...@gmail.com> wrote:

Dear Beast Users, 

I am trying to create an Extended Bayesian Skyline Plot from multiple loci data. The number of samples isnt great so I am trying to run my data for millions of chains which ends up with a super big log file. The problem is that I still get ESS low values around 160 in posterior, prior and meanRate. When setting the prior in Beauti I am giving a fixed value on the clock rate prior and a uniform distribution for the Demographic PopulationMean. I also tried to use the logCombiner to improve the ESS value but the resultant file is just too big!! I usually log every 1000 with a Length of chain of 100000000. Is anyone experiencing the same problem? Any suggestions? Thanks a lot!! 

--
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 https://groups.google.com/group/beast-users.
For more options, visit https://groups.google.com/d/optout.

[uxia saraiba esperon moldes]

I tried with combining 5 independent runs and then reducing the resampling states at lower frequency with 10000 in the logcombiner and it just got worse the ESS values now are poorer...am I doing something wrong? Thanks a lot for your quick response!

[uxia saraiba esperon moldes]

I have another doubt, I am using the logcombiner in order to increase the quality of my .log file but as a result I get another log file which I cannot use to create my Skyline plot because I am using the Extended Skyline Plot as a tree prior so I cant use Tracer to form the plot. I am currently using the csv file created by Beast  to draw the plot with a script designed in R but I cant get this csv file from logCombiner so how can I proceed?