search

NBISweden / MrBayes

MrBayes is a program for Bayesian inference and model choice across a wide range of phylogenetic and evolutionary models. For documentation and downloading the program, please see the home page:
http://NBISweden.github.io/MrBayes/
GNU General Public License v3.0
234 stars 79 forks source link

Partitioned data with starting tree #110

Open pbravakos opened 5 years ago

pbravakos commented 5 years ago

What is the issue that you are having?

I am running MrBayes in batch mode with partitioned data and a starting tree which are read correctly, but it gets stuck in the output bellow: 

 Initial log likelihoods and log prior probs for run 1:
      Chain 1 -- -179769313486231570814527423731704356798070567525844996598917476803157260780028538760589558632766878171540458953514382464234321326889464182768467546703537516986049910576551282076245490090389328944075868508455133942304583236903222948165808559332123348274797826204144723168738177180919299881250404026184124858368.000000 -- -nan

   There are 11 more chains on other processor(s)

   Using a relative burnin of 25.0 % for diagnostics

   Chain results (90000000 generations requested):

      0 -- [-179769313486231570814527423731704356798070567525844996598917476803157260780028538760589558632766878171540458953514382464234321326889464182768467546703537516986049910576551282076245490090389328944075868508455133942304583236903222948165808559332123348274797826204144723168738177180919299881250404026184124858368.000] [...11 remote chains...]

What is the environment that you run MrBayes in?



### Other information that may be of use to us in resolving this issue
It seems to be running, and makes use of all available cpus, but no output is generated, even if i wait many hours. If i run the exact same command without a starting tree then i get a normal output (and a log likelihood, instead of "nan").
Thanks 
Panos
nylander commented 5 years ago

Dear pbravakos, thank you for your report. Indeed, the output indicates an erroneous program behavior. We would, however, be helped by your exact input commands, and preferably also your input data (not shared by anyone else), in order to reproduce this issue. Please contact us personally if you wish to share your information. Cheers Johan

pbravakos commented 5 years ago

Dear Johan, I sent you an email with the input data and commands. Hope you recieved it. Cheers Panos

nylander commented 5 years ago

Dear Panos, Received your email. Thank you so much. /Johan

Den mån 17 juni 2019 17:52pbravakos notifications@github.com skrev:

Dear Johan, I sent you an email with the input data and commands. Hope you recieved it. Cheers Panos

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/NBISweden/MrBayes/issues/110?email_source=notifications&email_token=AANEKMO7HTY6ELRERHFHMUTP26XMBA5CNFSM4HYGJYZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODX3TRZI#issuecomment-502741221, or mute the thread https://github.com/notifications/unsubscribe-auth/AANEKMPL7NJZYY4ZCU6P65TP26XMBANCNFSM4HYGJYZA .

nylander commented 5 years ago

gh-110

Description

When analyzing partitioned protein data with a mixed aa model, likelihood calculations will fail unless one unlinks the aamodel. See issue #112.

In addition, user tree contain zero-length branches, which, in general, poses problems when used in optimization. With this example data, the program freezes after N generations.

Workaround

Original commands, with workarounds added

begin mrbayes;
    set autoclose=yes nowarn=yes;
    exe Pseudomonas_MSA_MrBayes_no_zero_length_branches.nex;
    lset applyto=(all) nucmodel=protein rates=invgamma ngammacat=4;
    unlink statefreq=(all) revmat=(all) shape=(all) pinvar=(all) aamodel=(all);
    prset applyto=(all) ratepr=variable aamodelpr=mixed topologypr=uniform;
    report siterates=yes;
    startvals tau=PhyMLTree V=PhyMLTree;
    mcmcp filename=Pseudomonas_MrBayes ngen=100000 nruns=2 nchains=4 relburnin=yes burninfrac=0.25 mcmcdiagn=yes;
    mcmcp stoprule=yes stopval=0.01 savetrees=yes Checkpoint=yes Checkfreq=5000 Printfreq=10;
    mcmcp nperts=6;
    mcmc;
    sumt contype=Halfcompat ntrees=1 conformat=figtree Showtreeprobs=No;
    sump;
    quit;
end;

Comments

nylander commented 5 years ago

Closing this and referring to issue #112

pbravakos commented 5 years ago

Dear Johan, Thanks for the immediate response. I followed the instructions (both for the tree in the nexus file and the MrBayes command) but I think I still have a problem. Now, I get a normal log likelihood but it seems that the analysis stagnates after a few generations. My output now is:

Initial log likelihoods and log prior probs for run 1:
      Chain 1 -- -14244.398223 -- 67.701678

   There are 11 more chains on other processor(s)

   Using a relative burnin of 25.0 % for diagnostics

   Chain results (90000000 generations requested):

      0 -- [-14244.398] [...11 remote chains...]
   1000 -- (-10163.815) [...11 remote chains...] -- 1774:58:48
   2000 -- (-9254.135) [...11 remote chains...] -- 2162:27:06
   3000 -- (-9138.519) [...11 remote chains...] -- 2374:55:15
   4000 -- (-9109.190) [...11 remote chains...] -- 2524:53:16
   5000 -- (-9096.552) [...11 remote chains...] -- 2649:51:10

   Average standard deviation of split frequencies: 0.123913

   6000 -- (-9102.142) [...11 remote chains...] -- 2828:58:41

No matter how long i wait, no more generations seem to be calculated. The number of generations at which the analysis seems to get stuck is not something constant, but can change. I was wondering if this is a problem of my system or a bug of the program. Thanks Panos

nylander commented 5 years ago

Dear Panos, I'm assuming this is the bug. We have to await the development in issue #112 /Johan

ronquist commented 4 years ago

Issue 112 fixed but it seems like there is an additional issue with the run hanging after some generations with this dataset. I will look into this.