SNAPP low ESS

[Lauren White]

Hi All,

I'm trying to run SNAPP for the first time and just have some queries about how it is running.

Firstly it is very slow, it's taken about 36 hours to run 100000 generations. Is this to be expected? I have read that SNAPP is computationally heavy.

I'm using 10 individuals from 4 populations with ~5000 SNPs. The 10 individuals are a subset of a larger dataset of individuals.

Secondly the posterior ESS doesn't seem to be going anywhere. For those first 100000 runs it has jumped around below 10, but doesn't seem to be increasing.

Is this normal? Is there something I can do about this?

Thank you for your help!

-Lauren

[Remco Bouckaert]

Hi Lauren,

If you are using 10 diploid individuals from 4 populations, there are 80 lineages and when using a single thread it is possible you get such long computation times. 

Did you experiment with the number of threads to see where SNAPP performs best? In general, more threads lead to better performance, but only up to a point where overhead results in diminishing returns.

If the sequences are diploid, you might want to try 5 individuals per population first, which should give you a sense whether there is signal in the data and it should run much faster.

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 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.