Hi Johannes. I will try to help you out, until Edward returns.
As a disclaimer, I can say that I never performed model-free analysis, and Edward is the expert. So, if he says something different, he is right. ;) I have been performing a long range of dispersion analysis, so I know some of the underlying parts of relax. You have 2 options. Either you could try to establish yourself with using multiple cores for calculation. I remember Edward told me, that model-free analysis calculations can run for 2-3 weeks. So if you could get access to a computer with 8, 20 or more cores, then your computation time will drop. It will probably take you 2-3 days, depending on your computer skills, to get the mpirun setup up and running. Or look here: http://wiki.nmr-relax.com/Run_relax_at_Google_Cloud_Computing http://wiki.nmr-relax.com/OpenMPI Another option is to allow for a more "sloppy" result. Change number of Monte-Carlo simulations from 500 to 10. You will still get the "minimum" right, but your error estimation of your result will be "sloppy". For publication, normally 500 to 2000 MC is fine. But for "searching", just use 10. Another possibility, you can actually track down why relax take so long time. Have a look in: /relax_dir/auto_analyses/dauvergne_protocol.py This is the line: ------ # Some class variables. opt_func_tol = 1e-25 opt_max_iterations = int(1e7) ------ relax will keep searching for a minimum for either maximum 10 million iterations, or if the difference of chi2 between an two iterations is lower than 1e-25. This is quite a strict rule. Really strict. It is "hard-coded" in the file, and it is normally not meant to be changed. But you could try setting: opt_func_tol = 1e-7 And then work "upwards", until you finish in "some reasonable time". As a reference: http://docs.scipy.org/doc/scipy/reference/optimize.linprog-simplex.html#optimize-linprog-simplex ..../python2.7/site-packages/scipy/optimize/optimize.py Here "ftol" is standard: 1e-4 So, relax is extremely strict. (Maybe too strict). How can you modify this? Either you could change the hardcoded value. Or you could use scripts to setup the calculations. Inspiration for this would be: /relax_dir/test_suite/system_tests/scripts/model_free/dauvergne_protocol.py And even more can be found here: /relax_dir/test_suite/system_tests/model_free.py (Note, that stuff like "self.interpreter." and so on should be deleted, and such".) (This is the full execution of all kinds of systemtests, which make sure that we know model-free is performing as it should) Please respond by using, "reply all", so other users can find this information on the mailing list. I will be happy to try answering any other questions. Best Troels Schwarz-Linnet PhD student, SBiNLab, Copenhagen University 2016-08-03 18:17 GMT+02:00 Johannes Dietschreit <dietschr...@gmail.com>: > Hello, > > I am trying to analyze the NMR relaxation data of a protein with respect > to the model-free analysis. I have two sets of R1, R2 and NOE data measured > at different field strengths. I set up the analysis with the > dauvergne_protocol.py I chose all models in the spirit of the fully > automated analysis. The first few diffusion models were computed relatively > fast but the program has been on the ellipsoid model for days. > The local_tm was run first, followed by sphere(1 round) , prolate(1024 > rounds), oblate (906 rounds). However, the ellipsoid model is at round > 22342. Can I check somehow whether the program is converging every so > slowly or whether I should abort the calculation? > > Thanks you for the help! > > Sincerely, > > Johannes Dietschreit > > > _______________________ > Johannes C. B. Dietschreit > BSc Science > _______________________________________________ > relax (http://www.nmr-relax.com) > > This is the relax-users mailing list > relax-users@gna.org > > To unsubscribe from this list, get a password > reminder, or change your subscription options, > visit the list information page at > https://mail.gna.org/listinfo/relax-users > _______________________________________________ relax (http://www.nmr-relax.com) This is the relax-users mailing list relax-users@gna.org To unsubscribe from this list, get a password reminder, or change your subscription options, visit the list information page at https://mail.gna.org/listinfo/relax-users
