reco::PFTau::signalTauChargedHadronCandidates() returns empty vector

[veelken]

Hi,

I came across a weird feature today: I produce a collection of reco::PFTau objects in a 1st cmsRun job and then read the same reco::PFTau collection in a 2nd cmsRun job.
While the function
reco::PFTau::signalTauChargedHadronCandidates().size()
returns values of 1 for one-prong taus and 3 for three-prong taus in the 1st cmsRun job as expected,
when I call the same function in the 2nd cmsRun job, I always get 0, i.e. the function
https://github.com/cms-sw/cmssw/blob/master/DataFormats/TauReco/interface/PFTau.h#L155
returns an empty vector in case the collection of reco::PFTau objects is read from a ROOT file.
I do not get a warning or error message. The code just fails silently.

The weird thing is that calling the function
reco::PFTau::signalPiZeroCandidates().size()
returns the same value in the 1st and in the 2nd cmsRun job.

Is the feature that I observed for the access to reco::PFTau::signalTauChargedHadronCandidates() the intended behaviour ?

The feature prevents the tau decay mode discriminator to be recomputed, because the tau mass window cuts cannot be applied (when reco::PFTau::signalTauChargedHadronCandidates().size() returns 0, this line of code fails: https://github.com/cms-sw/cmssw/blob/master/RecoTauTag/RecoTau/plugins/PFRecoTauDiscriminationByHPSSelection.cc#L136-L137 )

[cmsbuild]

A new Issue was created by @veelken Christian Veelken.

@Dr15Jones, @dpiparo, @silviodonato, @smuzaffar, @makortel, @qliphy can you please review it and eventually sign/assign? Thanks.

cms-bot commands are listed here

[slava77]

assign reconstruction

@swozniewski @mbluj
(in case you did not know of this already by other means)

[cmsbuild]

New categories assigned: reconstruction

@slava77,@perrotta,@jpata you have been requested to review this Pull request/Issue and eventually sign? Thanks

[slava77]

assign core

this may be connected to root

[cmsbuild]

New categories assigned: core

@Dr15Jones,@smuzaffar,@makortel you have been requested to review this Pull request/Issue and eventually sign? Thanks

[makortel]

@veelken Please provide a reproducible example

[veelken]

Hi,

I send you instructions for reproducing the problem attached below [*],
together with the output that I get when executing these commands on my machine [**].

The output that you get by following the instructions [*] corresponds to the case of the 2nd cmsRun job that I mentioned when I opened the issue.

There are two notable things in the output:

1. the function tauRef->signalTauChargedHadronCandidates().size() that is called in DumpRecoPFTaus.cc always returns zero, even though the taus are reconstructed in the one-prong decay mode (decayMode = 0) or in the three-prong decay mode (decayMode = 10 or 11)
2. the value of the recomputed decayModeFindingNewDMs tau ID discriminator is different from the original value of this discriminator (the difference is due to the fact that no tau mass window cuts are defined for the case that tauRef->signalTauChargedHadronCandidates().size() is equal to zero)

With kind regards,

Christian

[*]

mkdir debugCMSSW
cd debugCMSSW
cmsrel CMSSW_11_1_0
cd CMSSW_11_1_0/src/
cmsenv
mkdir -p HLTrigger/TallinnHLTPFTauAnalyzer
cd HLTrigger/TallinnHLTPFTauAnalyzer
cp -r /afs/cern.ch/user/v/veelken/public/debugCMSSW/CMSSW_11_1_0/src/HLTrigger/TallinnHLTPFTauAnalyzer/* .
scram b -j 4
cd test
cmsRun debugPFTauChargedHadrons_cfg.py

[**]

Begin processing the 1st record. Run 1, Event 294229, LumiSection 2073 on stream 0 at 24-Jul-2020 18:24:08.497 EEST
<DumpRecoPFTaus::analyze (moduleLabel = dumpSelectedHLTHpsPFTaus8HitsMaxDeltaZWithOfflineVertices)>:
src = InputTag: label = hltSelectedHpsPFTaus8HitsMaxDeltaZWithOfflineVertices, instance =
PFTau #0: pT = 34.7, eta = 0.433256, phi = 0.188892, decayMode = 0 (nCharged = 0, nPiZeros = 0), mass = 0.13957
chargedIso = 4.17266
lead. ChargedPFCand: pT = 34.7, eta = 0.433256, phi = 0.188892
lead. Track: pT = 34.7, eta = 0.433256, phi = 0.188892
discriminators:
hltHpsPFTauDiscriminationByDecayModeFindingNewDMs8HitsMaxDeltaZWithOfflineVertices = 1
hltHpsPFTauDiscriminationByDecayModeFindingNewDMs8HitsMaxDeltaZWithOfflineVerticesRECOMPUTED = 0
PFTau #1: pT = 91.1022, eta = -0.808807, phi = 1.06213, decayMode = 11 (nCharged = 0, nPiZeros = 1), mass = 1.56799
chargedIso = 0
lead. ChargedPFCand: pT = 26.6917, eta = -0.812224, phi = 1.05287
lead. Track: pT = 26.6917, eta = -0.812224, phi = 1.05287
discriminators:
hltHpsPFTauDiscriminationByDecayModeFindingNewDMs8HitsMaxDeltaZWithOfflineVertices = 1
hltHpsPFTauDiscriminationByDecayModeFindingNewDMs8HitsMaxDeltaZWithOfflineVerticesRECOMPUTED = 0
PFTau #2: pT = 39.9748, eta = 1.10178, phi = -2.39523, decayMode = 10 (nCharged = 0, nPiZeros = 0), mass = 1.37455
chargedIso = 22.6025
lead. ChargedPFCand: pT = 17.093, eta = 1.13523, phi = -2.39882
lead. Track: pT = 17.093, eta = 1.13523, phi = -2.39882
discriminators:
hltHpsPFTauDiscriminationByDecayModeFindingNewDMs8HitsMaxDeltaZWithOfflineVertices = 1
hltHpsPFTauDiscriminationByDecayModeFindingNewDMs8HitsMaxDeltaZWithOfflineVerticesRECOMPUTED = 0
...

[mbluj]

assign reconstruction

@swozniewski @mbluj
(in case you did not know of this already by other means)

Thanks for this notification. Unfortunately, I am at holidays until 3 August and because of this I will not be able to take a look sooner.

[makortel]

@veelken Thanks for the recipe, but this step does not work

cp -r /home/veelken/debugCMSSW/CMSSW_11_1_0/src/HLTrigger/TallinnHLTPFTauAnalyzer/* .

because I don't have access to your /home/veelken. Could you put that e.g. on AFS or as a tarball somewhere to download? Thanks.

[veelken]

@makortel I have copied the files to /afs/cern.ch/user/v/veelken/public/debugCMSSW/CMSSW_11_1_0/src/HLTrigger/TallinnHLTPFTauAnalyzer and updated the instructions above accordingly. Sorry for not realizing the issue with the file access earlier.