CompEvol
Wrong posterior and likelihood values
To whom it may concern
I'm trying to run BEAST 2.6.3 on this linguistic data set, using Relaxed Clock Log Normal. Tracer shows that posterior and likelihood have some reasonable mean and ESS values for the first few million chain runs and then the values get very large (e.g. 4.536E248) and ESS stops showing any values no matter how long the chain is. Every time I rerun the same .xml file, tree.Likelihood parameters fail for a few different partitionings (i.e. Tracer shows large mean values and no ESS values for one or two linguistic meaning classes). I've come across similar complaints and possible solutions here. However, I've not been able to solve this issue with my data. So, could you please help me with what exactly to be done with my data/parameters?
Best regards, Lexo
Can you post a screenshot from Tracer? Do any of the other parameter values get very large or very small at the same time that the posterior climbs to high numbers?
At first glance, you're asking a lot of a very small dataset -- one partition per word, with a rate per partition, on 7 taxa. It's not going to work. Combine all the partitions and try again.
Dear Alexei Thanks for your reply. Find attached the screenshots and the end lines of the output report. Up to run 30,000,000 all seemed fine. Best regards,Lexo On Wednesday, May 12, 2021, 02:13:19 AM GMT+5, Alexei Drummond @.***> wrote:
Can you post a screenshot from Tracer? Do any of the other parameter values get very large or very small at the same time that the posterior climbs to high numbers?
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
Dear Simon Thanks for your response. The thing is it works fine about half way through a 10^8-long chain, after which mixing begins to fail. Best regards,Lexo On Wednesday, May 12, 2021, 02:23:17 AM GMT+5, Simon J Greenhill @.***> wrote:
At first glance, you're asking a lot of a very small dataset -- one partition per word, with a rate per partition, on 7 taxa. It's not going to work. Combine all the partitions and try again.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.