Difficulty in getting convergence for some parameters

[Kim]

Hi, I'm having difficulty in getting some parameters to converge for a migrate run with 3 populations.  Attached are output files from the last 2 runs I've done.  It seems like most of the parameters have converged, but there still seems to be a lot of fluctuation for parameters 7 and 9.  I've tried increasing sampling but am hesitant to increase further because the runs already take so long to complete.  I've been running migrate-n-mpi on only 3 nodes, in order to leave enough nodes available for other users on the system.  Any suggestions besides increasing sampling?

Thanks,

Kim

[Peter Beerli]

I would call it a day and simply state that those parameters may or may have not converged ,given that the marginal likelihoods are  similar and your results look also rather similar between the two runs, a longer run will not change the results (for mode and credibility set), It seems that your data is highly variable given the relatively high values for two of the thetas, if you think that site rate variation may be an issue you could run a short testrun by using a F84 + G as the mutation model (but you will need to get the alpha parameter of the Gamma distribution from somewhere else).

It seems that your 9 parameter model is on the upper limit for your data given that some of histograms have very heavy tails and may be dominated by the prior for large values of the thetas.

peter

--
You received this message because you are subscribed to the Google Groups "migrate-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to migrate-suppo...@googlegroups.com.
To post to this group, send email to migrate...@googlegroups.com.
Visit this group at http://groups.google.com/group/migrate-support.
For more options, visit https://groups.google.com/d/optout.
<Cverater_concatenated_horizontal_Migrate10.pdf><Cverater_concatenated_horizontal_Migrate9.pdf>

[Kim]

Thanks for the input, Peter.  I've had similar results with a few other runs, so it seems like I probably should stop trying to optimize those as well.