cms-nanoAOD / cmssw

CMS NanoAOD software integration repository

http://cms-sw.github.io/

Apache License 2.0

3 stars 10 forks source link

Nan with quark-gluon in NanoAOD #556

Closed mariadalfonso closed 3 years ago

mariadalfonso commented 3 years ago

reported here https://indico.cern.ch/event/962278/#7-issues-with-quark-gluon-in-n

mariadalfonso commented 3 years ago

From @alefisico I think this is related with a mismatch of QGL databases in the global tags reported to us last week. To summarize: 2018UL MC GT: QGLikelihoodObject_v1_AK4PFchs (2015 training) 2017UL MC GT: QGLikelihoodObject_v1_AK4PFchs_2017 (2017 training) 2016UL MC GT(preVFP, postVFP): QGLikelihoodObject_v1_AK4PFchs_2016 (2016 training) Data GT: QGLikelihoodObject_v1_AK4PFchs (2015 training) — the same GT is used for all three years,

It should be the 2017 training everywhere (until we have dedicated UL trainings).

As extra information, we know that the 2015 trainings can produced these Nan-values.

hqucms commented 3 years ago

FYI -- We have requested new GTs from AlCa/DB to fix this: https://hypernews.cern.ch/HyperNews/CMS/get/calibrations/4357.html

mariadalfonso commented 3 years ago

added in the GT with PR

32247

mariadalfonso commented 3 years ago

I still see a lot of nan for nanoV8 campaign

here is some way to reproduce them

root://xrootd-cms.infn.it//store/mc/RunIISummer20UL18NanoAODv2/TTToSemiLeptonic_TuneCP5_13TeV-powheg-pythia8/NANOAODSIM/106X_upgrade2018_realistic_v15_L1v1-v1/20000/A2B51438-22E7-C440-9490-B75A84BBBB3F.root

FOUND nan in Jet_qgl[0] run:lumi:event 1:132039:132038327 FOUND nan in Jet_qgl0] run:lumi:event 1:341110:341109793 FOUND nan in Jet_qgl[0] run:lumi:event 1:130399:130398301 ...

alefisico commented 3 years ago

Hi, sorry for the late reply. We dont rerun QGL in nanoAOD, we take the value directly from miniAOD. The nanoAODv8 file that you posted is done with the global tag where the correct QGL training is included but the miniAOD parent is done with an older global tag (106X_upgrade2018_realistic_v11_L1v1). This problem is solved if the re-miniAOD campaing was done with the 106X_upgrade2018_realistic_v15_L1v1 global tag.

alefisico commented 3 years ago

Hi @mariadalfonso after further checks and discussions about this issue, it looks like what I said before is not true. We are looking into ways of fixing this issue asap. Maybe you should open this issue again.

mariadalfonso commented 3 years ago

a protection being added in

https://github.com/cms-sw/cmssw/pull/33925/files#diff-377770d6a7d4705da48aa96709bdedefcb42d5b5497c85ba41a3f11074c148a0R243