Back to TotalFit Methods

Monitoring fitting progress

It is instructive to examine success of a randomized fitting run by observing fitting results of EVERY monte-carlo trial. To do this use single-core mode and set monte_carlo_monitoring='SEE-ALL-FIGURES' on a single core and you will see figures for every individual monte-carlo fitting run. Starting conditions and final results are output on screen, the best-fit figure is plotted and the program stops in a 'keyboard' mode. At this time you can examine any variables you like though command line or a Workspace. To continue with the next run type 'return' on MATLAB command line, and to completely stop and exit: 'dbquit'.

Turn monte_carlo_monitoring back to OFF or BRIEF before running real multi-core fitting run (figures slow calculations down GREATLY). Setting to 'BRIEF' only outputs SS and exitflag of every trial. NOTE that if run on multiple cores on remote computer the 'SEE-ALL-FIGURES' mode may lead to crash of remote deamons because they will not have ability to display graphics!

IMPORTANT NOTE: The key assumption of the Monte-Carlo procedure is that randomization of the ideal (simulated) data reproduces data scatter very similar to experimentally observed. If this is not the case it means two important things:

(1) you do not appropriately estimate uncertainties in your data

(2) Monte-Carlo confidence intervals obtained with the current uncertainties and randomization are MEANINGLESS. In this case---stop and better investigate sources of experimental error in your measurements.



For a specific usage see TotalFit.m