How to continue a interrupted MrBayes tree analysis?

Question

[超]

Hello!

My MrBayes analysis stopped due to timeout in HPC. I tried add append=yes in nex file to continue the rest analysis.

begin mrbayes;
mcmc nchains=16 ngen=10000000 append=yes stoprule=yes stopval=0.1;
sump;
sumt;
end;

But when I execute nex file, errors come.

Chain 16 -- -419920.472208 -- -1624.922043
Could not rename file inputfile.nex.run1.t
Error appending to previous run
Could not rename file inputfile.nex.run2.t
Error appending to previous run

And this is sbatch file.

# load mpirun.
module load openmpi-1.8/intel
mpirun -np 32 /scratch/bio/bin/mb  inputfile.nex >inputfile.logs

How can I solve this problem?

[omaridris5315]

Hello Mensur, Could you look at my case too? interruption issue but a different error. This is the second time I face and I rerun it. I have a pc and it is slow as I am not yet sure how to use gpu or mpi option to start over, I have wasted 20 days between two runs.

Thankyou.

Answer 1

[Mensur Dlakic]

What you did was correct. However, by sheer bad luck the writing process for one of the chain traces may have been interrupted mid-way through, in which case you could be starting from an incomplete trace. If that's the case, I don't think there is a way to recover.

You can test this yourself by manually stopping the run, but make sure to do it in between chain trace writing intervals. If you can recover those runs, then the reason you can't do the same with this run is because of incomplete trace.

[超]

Thank you very much, Mensur! I will try to do the test or just rerun this analysis. Here, I also have some other information to report. When I execute new nex file with "append=yes", mcmc run1.p run1.t run2.p run2.t file all change to a empty file or only end is in these files like below pictures.

These are files before executing nex file with append=yes. I don't know if that happens in normal operation or some error in my process.

[Mensur Dlakic]

All your trace files before continuation seem to have the same sizes (run1.p vs run2.p or run1.t vs run2.t), so it is less likely that my hypothesis about incomplete trace files is correct. You may need to get some advice from your system admins or directly from MrBayes authors.