MCMCTree analysis not starting

[Miles Zhang]

Hi I am trying to run a MCMCTree analysis for a big UCE dataset of 292 tips and 100 loci, however I don't think it is running despite the job not being killed on HPC. 

The output's last line are:

Homogeneity statistic: X2 = 0.89317 G = 0.89848

Average                        0.26072 0.22070 0.28007 0.23851
(Ambiguity characters are used to calculate freqs.)


# constant sites:      0 (0.00%)

Whereas the log file gets stuck after spitting out 2.5 newick trees, with the third one being a partial (see attached).

I've tried doing this with both sequence and approximation and neither is working. I assume it is some kind of error in my input files but I am not sure how to debug it.

Thanks,

Miles

[Sishuo Wang]

Dear Yuan,

How much time did you spend before killing it? Seems to me that the program is still there which means it could be possible that some time later you'll see more outputs gradually coming out. Also, could you pls have a try with a shorter alignment say 100 sites? The reason is bc your alignment has some 47k sites and 292 tips. Thx!

best,

sishuo

[Miles Zhang]

Hi Sishuo,

  I left it running for a week and nothing happened, I also tried 100 sites and it was stuck in the same area.

Thanks,

Miles

[Sishuo Wang]

Dear Miles,

I c. I think your burnin = 5000000 is way too large. I ran with your ctl with usedata=3 and burnin=50, samfreq=50, nsample=10, the result shown in the screenshot, but note that these values are just to test if the program runs on your data rather than saying that those values are "good" enough.

best,

sishuo

[Miles Zhang]

Thank you Sishuo, I did the same test and indeed it worked so I will try reducing the burnin in my actual analysis. It seems like if I run this locally it will reach the "starting mcmc" step, but gets stuck on the same area as before if I do it on HPC even with a much reduced data set. Do you happen to know if this is a memory allocation issue or something else?

Thanks,

Miles

[Sishuo Wang]

Dear Miles,

Could that be bc on hpc the output may be stored before being presented to the user? I am not very sure actually. You might also wish to try some examples using the files given by paml. Or, you can set usedata=0 to see if there's any result coming out when run on hpc as that means sampling from the effective priors so should be super fast otherwise there might be sth that goes wrong.

thx,

sishuo