search

cms-analysis / HiggsAnalysis-CombinedLimit

CMS Higgs Combination toolkit.
https://cms-analysis.github.io/HiggsAnalysis-CombinedLimit/latest
Apache License 2.0
75 stars 381 forks source link

prefit nuisance values in FitDiagnostics all 0±0 when lnU's present #600

Open adewit opened 4 years ago

adewit commented 4 years ago

This was not the case in older combine versions (tested comb2017 branch in CMSSW_8_1_0). Making this issue because it's not clear to me exactly why this is happening and I don't have time to continue investigating now.

VourMa commented 4 years ago

Hi all, I am facing the same issue. I have setup the recommended version following the instructions here and I have verified that the existence of lnU nuisance parameters causes the prefit NP values to be 0±0. I was wondering whether there has been any progress in the investigation. Thanks a lot in advance!

adewit commented 4 years ago

No, not yet. lnU's are not so widely used now (because rateParams effectively do the same thing, but are easier to interpret), so it hasn't come up again in the past months and I forgot about it.

VourMa commented 4 years ago

I noticed this issue because it also breaks the diffNuisance result. However, I can get the correct pulls when computing the impacts, so it's not that big of a problem. What I am more worried about is that this could affect other results without me noticing. About the rateParams, thank you for pointing them to me. I have been doing some quick tests using these instead of the lnUs and the results seem pretty similar. Is the recommended solution to replace lnU NP with the appropriate rateParam ones?

adewit commented 4 years ago

From what I remember this doesn't actually affect any of the fits/results (most things don't rely on the pre-fit values of the NP's), so just diffNuisances. That said if you're happy to work with rateParams it's certainly a good idea to switch to those

VourMa commented 4 years ago

Great, thanks a lot for the clarifications!