URL: <http://gna.org/bugs/?7845> Summary: full_analysis.py problem with the chi2 from the previous run (precision too high ?) Project: relax Submitted by: semor Submitted on: Friday 11/24/2006 at 16:22 Category: None Severity: 3 - Normal Priority: 5 - Normal Status: None Privacy: Public Assigned to: None Originator Name: Originator Email: Open/Closed: Open Discussion Lock: Any Release: Repository: 1.2 line Operating System: GNU/Linux _______________________________________________________ Details:
From the 1.2 line at revision 2883...
The new full_analysis.py script has the ability to tell if convergence has been reached by looking at the chi2, model-free models, diffusion tensor parameters and model-free parameters from the previous round of optimization. This should prevent running the final global optimization before every diffusion tensor convergence. However, the chi2 test doesn't work. After a lot of optimization rounds with the old script, I tried the new one. For example, it was the 12th round for the prolate diffusion tensor. The first time I used the new script, it told me the previous chi2 was 7086.7226167700001 and the new one was 7086.7226167680947. I then continued and got the same exact chi2 for the previous and new round... I did it another time and got the exact same thing again. So, the 'previous' chi2 is always 7086.7226167700001 ans the 'present' chi2 is always 7086.7226167680947 for this particular diffusion tensor using my data. I get the same behaviour for every diffusion tensor, so this is global error. Also, the chi2 difference is extremely small, not even visible when taking a look at the results.bz2 file, since the round up values are the same. The other tests (model-free models, diffusion tensor parameters and model-free parameters) seem to work. Maybe this is related to the calculation precision... Thanks for help ! Séb _______________________________________________________ Reply to this item at: <http://gna.org/bugs/?7845> _______________________________________________ Message sent via/by Gna! http://gna.org/