Opened 7 years ago
Closed 7 years ago
#76 closed defect (invalid)
WARNING in LauAbsFitModel::getLogLikelihood : Strange likelihood value
| Reported by: | guest | Owned by: | |
|---|---|---|---|
| Priority: | major | Milestone: | |
| Version: | Keywords: | ||
| Cc: |
Description
When I am performing toyMC test with a large amount of sample, I found that a large fraction of the fit would show the error message: Returning worst NLL found so far to force MINUIT out of this region. WARNING in LauAbsFitModel::getLogLikelihood : Strange likelihood value ... It looks like an events has strange likelihood value with the parameter set in a iteration. In the fit result, the floated parameters would be forced as -256768, then it would be stopped. For RooFit, if the returned likelihood value is strange (e.h. NAN), the fitter would try to change the value in the next iteration.
In Laura++, is there a parameter to change the setup so that the fitter can keep trying adjusting the parameters even if the error happens rather than directly stopping the fit?
Thank you.
Laura++ will try to continue the fit by forcing MINUIT out of the unphysical region by returning the worst likelihood value it has so far encountered. However, this will not always be successful. It depends greatly on what exactly is causing the problem in the first place. Since this does not appear to be a bug I will close this ticket. I suggest instead that you send an email to the developers mailing list (see the laura.hepforge.org webpage for the address) attaching or linking to code that will reproduce the problem you are experiencing. We might then be able to advise you how best to proceed, e.g. modifying parameter limits or step sizes.
In general, contacting the mailing list is probably the best thing to do initially. If we then think you've found a genuine bug or you have a new feature request then we will ask you to create a corresponding ticket so that we can track the issue or schedule development of the new feature for a particular upcoming release.