Jet identification in high pile-up environment
Introduction
The identification of jets from pile-up relies on three types of propeties of the jets:- inside the tracker acceptance, the trajectories of tracks associated to the jets can be used to establish the compatibility of the jet with the primary interaction vertex;
- the topology of the jet shape can be used in order to disentangle jets arising from the overlap of multiple interactions from truly hard jets.
- the object mujltiplicity can be used as an additional handle.
| beta | |
| RMS (or dR2Mean) | |
| jet dR profile | |
| charged and neutral multiplicities |
- a cut-based selection explits the minimal amount of information in the simplest way;
- a more sophisticated approach combined all the discriminating variables in a Boosted Decision Tree. Two different BDTs are provided: * A first BDT, labelled as simple only explits beta and the jet profile.
- A second BDT, labelled full also integrates the information from the multiplicities.
Recommendation for Summer16MC:
The Summer16MC contains the latest training of the PU jet ID based on 80X.How to update JEC for PUjetID (prescription based on Moriond17MC and 2016 data):
The Moriond 2017 MC (CMSSW>=8_0_20) contains the latest training of the PU jet ID based on 80X. To rerun the PU jet ID on the 2016 data with the recommended JEC, cherry-pick the following commit into your CMSSW>=8_0_20 release:git remote add ahinzmann git@github.com:ahinzmann/cmssw.git git fetch ahinzmann PUidMiniAODfix80 git cherry-pick ca33756e1747aec27d13971bcfd0874b16724e7fAn example how to re-calculate the pileup jet ID and re-calibrate the jets on MiniAOD is below:
import FWCore.ParameterSet.Config as cms
process = cms.Process("PATUPDATE")
process.load("FWCore.MessageLogger.MessageLogger_cfi")
process.source = cms.Source("PoolSource",
fileNames = cms.untracked.vstring(["/RelValTTbar_13/CMSSW_8_0_5-
PU25ns_80X_mcRun2_asymptotic_2016_miniAODv2_v0_gs71xJecGT-v1/MINIAODSIM"])
)
process.maxEvents = cms.untracked.PSet( input = cms.untracked.int32(10000) )
process.load("Configuration.StandardSequences.FrontierConditions_GlobalTag_cff")
from Configuration.AlCa.GlobalTag import GlobalTag
process.GlobalTag = GlobalTag(process.GlobalTag, '80X_mcRun2_asymptotic_2016_miniAODv2')
process.load("RecoJets.JetProducers.PileupJetID_cfi")
process.pileupJetIdUpdated = process.pileupJetId.clone(
jets=cms.InputTag("slimmedJets"),
inputIsCorrected=True,
applyJec=True,
vertexes=cms.InputTag("offlineSlimmedPrimaryVertices")
)
print process.pileupJetId.dumpConfig()
process.load("PhysicsTools.PatAlgos.producersLayer1.jetUpdater_cff")
process.patJetCorrFactorsReapplyJEC = process.updatedPatJetCorrFactors.clone(
src = cms.InputTag("slimmedJets"),
levels = ['L1FastJet', 'L2Relative', 'L3Absolute'] )
process.updatedJets = process.updatedPatJets.clone(
jetSource = cms.InputTag("slimmedJets"),
jetCorrFactorsSource = cms.VInputTag(cms.InputTag("patJetCorrFactorsReapplyJEC"))
)
process.updatedJets.userData.userFloats.src += ['pileupJetIdUpdated:fullDiscriminant']
process.p = cms.Path( process.pileupJetIdUpdated + process.patJetCorrFactorsReapplyJEC + process. updatedJets )
process.out = cms.OutputModule("PoolOutputModule",
fileName = cms.untracked.string("patTupleUpdatedFromMiniAOD.root"),
outputCommands = cms.untracked.vstring('keep *')
)
process.endpath = cms.EndPath(process.out)
# import ROOT in batch mode
import sys
oldargv = sys.argv[:]
sys.argv = [ '-b-' ]
import ROOT
ROOT.gROOT.SetBatch(True)
sys.argv = oldargv
# load FWLite C++ libraries
ROOT.gSystem.Load("libFWCoreFWLite.so");
ROOT.gSystem.Load("libDataFormatsFWLite.so");
ROOT.AutoLibraryLoader.enable()
# load FWlite python libraries
from DataFormats.FWLite import Handle, Events
from PhysicsTools.HeppyCore.utils.deltar import deltaR
jets, jetLabel = Handle("std::vector<pat::Jet>"), "slimmedJets"
# open file (you can use 'edmFileUtil -d /store/whatever.root' to get the physical file name)
events = Events("file:////afs/cern.ch/user/h/hinzmann/workspace/tmp/MiniAOD.root")
for iev,event in enumerate(events):
#if iev >= 10: break
event.getByLabel(jetLabel, jets)
print "\nEvent %d: run %6d, lumi %4d, event %12d" % (iev,event.eventAuxiliary().run(), event.eventAuxiliary().luminosityBlock(),event.eventAuxiliary().event())
# Jets (standard AK4)
for i,j in enumerate(jets.product()):
if j.pt() < 20: continue
print "jet %3d: pt %5.1f (raw pt %5.1f, matched-calojet pt %5.1f), eta %+4.2f, pileup mva disc %+.2f" % (
i, j.pt(), j.pt()*j.jecFactor('Uncorrected'), j.userFloat("caloJetMap:pt"), j.eta(), j.userFloat("pileupJetIdUpdated:fullDiscriminant"))
print "Passes loose: ", bool(j.userInt("pileupJetIdUpdated:fullId") & (1 << 2)), "medium: ", bool(j.userInt("pileupJetIdUpdated:fullId") & (1 << 1)), "tight: ", bool(j.userInt("pileupJetIdUpdated:fullId") & (1 << 0))
. Validation on 2016 data is ongoing.
A preliminary training with 80X MC was also available as userfloat in 80X MiniAODv2, however, the new training described above is expected to give better overall performance.
The 80X MiniAODv1 contains a PU jet ID training based on 76X and you may want to rerun a new pileup jet ID training.
Information for 13TeV data analysis in 76X:
A preliminary 13 TeV training with 76X MC has been completed. Information on the performance can be found here: The recipe to install the corresponding code is below:cmsrel CMSSW_7_6_4 cd CMSSW_7_6_4/src/ cmsenv git cms-init git cms-merge-topic jbrands:pileupJetId76X git cherry-pick 7c23237a87181e9320874c6f1f913a43fe849499 cd RecoJets/JetProducers/data/ wget https://github.com/jbrands/RecoJets-JetProducers/raw/3dad903ed25d025f68be94d6f781ca957d6f86ac/pileupJetId_76x_Eta0to2p5_BDT.weights.xml.gz wget https://github.com/jbrands/RecoJets-JetProducers/raw/3dad903ed25d025f68be94d6f781ca957d6f86ac/pileupJetId_76x_Eta2p5to2p75_BDT.weights.xml.gz wget https://github.com/jbrands/RecoJets-JetProducers/raw/3dad903ed25d025f68be94d6f781ca957d6f86ac/pileupJetId_76x_Eta2p75to3_BDT.weights.xml.gz wget https://github.com/jbrands/RecoJets-JetProducers/raw/3dad903ed25d025f68be94d6f781ca957d6f86ac/pileupJetId_76x_Eta3to5_BDT.weights.xml.gz cd ../../.. scram b -j5Pileup jet ID calculator can recalibrate jets with JEC if necessary. Therefore it is important to load the correct global tag or DB file with the latest corrections (example below). An example how to calculate the pileup jet ID and re-calibrate the jets on MiniAOD is below:
import FWCore.ParameterSet.Config as cms
process = cms.Process("PATUPDATE")
process.load("FWCore.MessageLogger.MessageLogger_cfi")
process.source = cms.Source("PoolSource",
fileNames = cms.untracked.vstring(["/store/relval/CMSSW_7_6_2/RelValZMM_13/MINIAODSIM/PU25ns_76X_mcRun2_asymptotic_v12-v1/00000/C86BA73E-D09C-E511-AD68-002590596468.root"])
# fileNames = cms.untracked.vstring(["root://xrootd.unl.edu//store/relval/CMSSW_7_6_2/RelValTTbar_13/MINIAODSIM/PU25ns_76X_mcRun2_asymptotic_v12-v1/00000/DC900947-DC9C-E511-B3AC-0CC47A4C8F0A.root"])
)
process.maxEvents = cms.untracked.PSet( input = cms.untracked.int32(10000) )
process.load("Configuration.StandardSequences.FrontierConditions_GlobalTag_cff")
from Configuration.AlCa.GlobalTag import GlobalTag
process.GlobalTag = GlobalTag(process.GlobalTag, '76X_mcRun2_asymptotic_RunIIFall15DR76_v1')
process.load("RecoJets.JetProducers.PileupJetID_cfi")
process.pileupJetIdUpdated = process.pileupJetId.clone(
jets=cms.InputTag("slimmedJets"),
inputIsCorrected=True,
applyJec=True,
vertexes=cms.InputTag("offlineSlimmedPrimaryVertices")
)
print process.pileupJetId.dumpConfig()
process.load("PhysicsTools.PatAlgos.producersLayer1.jetUpdater_cff")
process.patJetCorrFactorsReapplyJEC = process.updatedPatJetCorrFactors.clone(
src = cms.InputTag("slimmedJets"),
levels = ['L1FastJet', 'L2Relative', 'L3Absolute'] )
process.updatedJets = process.updatedPatJets.clone(
jetSource = cms.InputTag("slimmedJets"),
jetCorrFactorsSource = cms.VInputTag(cms.InputTag("patJetCorrFactorsReapplyJEC"))
)
process.updatedJets.userData.userFloats.src += ['pileupJetIdUpdated:fullDiscriminant']
process.p = cms.Path( process.pileupJetIdUpdated + process.patJetCorrFactorsReapplyJEC + process. updatedJets )
process.out = cms.OutputModule("PoolOutputModule",
fileName = cms.untracked.string("patTupleUpdatedFromMiniAOD.root"),
outputCommands = cms.untracked.vstring('keep *')
)
process.endpath = cms.EndPath(process.out)
import FWCore.ParameterSet.Config as cms
process = cms.Process("PATUPDATE")
process.load("FWCore.MessageLogger.MessageLogger_cfi")
process.source = cms.Source("PoolSource",
fileNames = cms.untracked.vstring(["/store/relval/CMSSW_7_6_2/RelValZMM_13/GEN-SIM-RECO/PU25ns_76X_mcRun2_asymptotic_v12-v1/00000/56F71722-CC9C-E511-9FCF-0CC47A4D7668.root"])
)
process.maxEvents = cms.untracked.PSet( input = cms.untracked.int32(100) )
process.load("Configuration.StandardSequences.FrontierConditions_GlobalTag_cff")
from Configuration.AlCa.GlobalTag import GlobalTag
process.GlobalTag = GlobalTag(process.GlobalTag, '76X_mcRun2_asymptotic_RunIIFall15DR76_v1')
process.load("RecoJets.JetProducers.PileupJetID_cfi")
process.pileupJetId.jets=cms.InputTag("ak4PFJetsCHS")
process.pileupJetId.inputIsCorrected=False
process.pileupJetId.applyJec=True
process.pileupJetId.vertexes=cms.InputTag("offlinePrimaryVertices")
print process.pileupJetId.dumpConfig()
process.p = cms.Path(process.pileupJetId)
process.out = cms.OutputModule("PoolOutputModule",
fileName = cms.untracked.string("patTuplePuIdFromAOD.root"),
outputCommands = cms.untracked.vstring('keep *')
)
process.endpath = cms.EndPath(process.out)
# import ROOT in batch mode
import sys
oldargv = sys.argv[:]
sys.argv = [ '-b-' ]
import ROOT
ROOT.gROOT.SetBatch(True)
sys.argv = oldargv
# load FWLite C++ libraries
ROOT.gSystem.Load("libFWCoreFWLite.so");
ROOT.gSystem.Load("libDataFormatsFWLite.so");
ROOT.AutoLibraryLoader.enable()
# load FWlite python libraries
from DataFormats.FWLite import Handle, Events
from PhysicsTools.HeppyCore.utils.deltar import deltaR
jets, jetLabel = Handle("std::vector<pat::Jet>"), "updatedJets"
pujetids, pujetidLabel = Handle("edm::ValueMap<StoredPileupJetIdentifier>"), "pileupJetIdUpdated"
pujetidDiscriminant, pujetidDiscriminantLabel = Handle("edm::ValueMap<float>") , "pileupJetIdUpdated:fullDiscriminant"
pujetidFullId, pujetidFullIdLabel = Handle("edm::ValueMap<int>") , "pileupJetIdUpdated:fullId"
# open file (you can use 'edmFileUtil -d /store/whatever.root' to get the physical file name)
events = Events("file:////afs/cern.ch/user/h/hinzmann/workspace/tmp/patTupleUpdatedFromMiniAOD.root")
#eff_loose=0
#eff_medium=0
#eff_tight=0
#num_jets=0
for iev,event in enumerate(events):
#if iev >= 10: break
event.getByLabel(jetLabel, jets)
event.getByLabel(pujetidLabel, pujetids)
event.getByLabel(pujetidDiscriminantLabel, pujetidDiscriminant)
event.getByLabel(pujetidFullIdLabel, pujetidFullId)
print "\nEvent %d: run %6d, lumi %4d, event %12d" % (iev,event.eventAuxiliary().run(), event.eventAuxiliary().luminosityBlock(),event.eventAuxiliary().event())
# Jets (standard AK4)
for i,j in enumerate(jets.product()):
if j.pt() < 20: continue
print "jet %3d: pt %5.1f (raw pt %5.1f, matched-calojet pt %5.1f), eta %+4.2f, pileup mva disc %+.2f" % (
i, j.pt(), j.pt()*j.jecFactor('Uncorrected'), j.userFloat("caloJetMap:pt"), j.eta(), j.userFloat("pileupJetIdUpdated:fullDiscriminant"))
print "Passes loose: ", bool(pujetidFullId.product().get(i) & (1 << 2)), "medium: ", bool(pujetidFullId.product().get(i) & (1 << 1)), "tight: ", bool(pujetidFullId.product().get(i) & (1 << 0))
#string=""
#string+="MVA=%.3f, " % (pujetidDiscriminant.product().get(i))
#variables=['RMS', 'beta', 'betaClassic', 'betaStar', 'betaStarClassic', 'dR2Mean', 'dRMatch', 'dRMean', 'dZ', 'frac01', 'frac02', 'frac03', 'frac04', 'frac05', 'frac06', 'frac07', 'jetEta', 'jetPt', 'jetR', 'jetRchg', 'majW', 'minW', 'nCharged', 'nNeutrals', 'nParticles', 'nTrueInt', 'nvtx', 'ptD', 'pull', 'rho']
#for var in variables:
# string+=var+"=%.3f, " % (getattr(pujetids.product().get(i),var)())
#print string
#try: genJetMatch=deltaR(j.genJet().eta(),j.genJet().phi(),j.eta(),j.phi())<0.2 and ((abs(j.partonFlavour()) > 0 and abs(j.partonFlavour()) < 6) or (j.partonFlavour() == 21))
#except: genJetMatch=0
#if genJetMatch:
# num_jets+=1.
# eff_loose+=bool(pujetidFullId.product().get(i) & (1 << 2))
# eff_medium+=bool(pujetidFullId.product().get(i) & (1 << 1))
# eff_tight+=bool(pujetidFullId.product().get(i) & (1 << 0))
#print num_jets
#print eff_loose/num_jets
#print eff_medium/num_jets
#print eff_tight/num_jets
edm::Handle<edm::View<pat::Jet > >; jets;
iEvent.getByLabel("selectedPatJets",jets);
Handle<ValueMap<float> >; puJetIdMVA;
iEvent.getByLabel("pileupJetIdUpdated:fullDiscriminant",puJetMva);
Handle<ValueMap<int> >; puJetIdFlag;
iEvent.getByLabel("pileupJetIdUpdated:fullId",puJetMva);
for ( unsigned int i=0; i<jets->size(); ++i ) {
const pat::Jet & patjet = jets->at(i);
float mva = (*puJetIdMVA)[jets->refAt(i)];
int idflag = (*puJetIdFlag)[jets->refAt(i)];
cout << "jet " << i << " pt " << patjet.pt() << " eta " << patjet.eta() << " PU JetID MVA " << mva;
if( PileupJetIdentifier::passJetId( idflag, PileupJetIdentifier::kLoose ) {
cout << " pass loose wp";
}
if( PileupJetIdentifier::passJetId( idflag, PileupJetIdentifier::kMedium ) {
cout << " pass medium wp";
}
if( PileupJetIdentifier::passJetId( idflag, PileupJetIdentifier::kTight ) {
cout << " pass tight wp";
}
cout << endl;
}
Information for 8TeV data analysis:
The algorithms described above can be run using the CMSSW module available here.
The list of tags to be used with different CMSSW releases is reported in the table below.
The package provides three classes:
.
The list of tags to be used with different CMSSW releases is reported in the table below.
| CMSSW 4X | V00-03-04 | Automatically detects CMSSW version and used ultimate 4X weights when running on 4X |
| CMSSW 5X | V00-03-04 | Used for ICHEP analyses in 5X |
- the PileupJetIdentifier class (and the StoredPileupJetIdentifier subset) contains all the variables used for discrimination.
- the PileupJetIdAlgo class computes all variables in PileupJetIdentifier and as well as the jet identification MVAs.
- the PileupJetIdProducer class is a CMSSW producer that thakes collections of PAT jet in input and produces ValueMap with StoredPileupJetIdentifier, the final MVAs and flags with the result of the working points evaluation.
cvs co -r <TAG> -d CMGTools/External UserCode /CMG/CMGTools/External cd CMGTools/External scram b -j 5The procedure to run on PAT jets or plain reco jets is different, due to the different treatment of jet energy correction. In fact, while the input variables are independent of the jet energy corrections, the value of the final MVA depends on the corrected jet energy.
Running the producer on PAT jets
If you use PAT jets for your analysis applying the pile-up identification is as simple as running an additional sequence in you PAT process.# load the PU JetID sequence
process.load("CMGTools.External.pujetidsequence_cff")
# run the PU JetID sequence
process.p = cms.Path(process.patSequence * process.puJetIdSqeuence)
# keep the PU JetID products
process.out.extend(["keep *_puJetId_*_*", # input variables
"keep *_puJetMva_*_*" # final MVAs and working point flags
])
This will produce the following output:
edmDumpEventContent patTuple.root | grep puJet edm::ValueMap<StoredPileupJetIdentifier> "puJetId" "" "PAT" edm::ValueMap<float> "puJetMva" "fullDiscriminant" "PAT" edm::ValueMap<float> "puJetMva" "cutbasedDiscriminant" "PAT" edm::ValueMap<float> "puJetMva" "simpleDiscriminant" "PAT" edm::ValueMap<int> "puJetMva" "fullId" "PAT" edm::ValueMap<int> "puJetMva" "cutbased" "PAT" edm::ValueMap<int> "puJetMva" "simpleId" "PAT"For each type of algorithm (cut-based, simple and full BDTs), the producer outputs two ValueMap, one (*Discriminan) with the value of the MVA and a second (*Id) encoding the working point evaluation. This information can be accessed at analysis level in the following way:
edm::Handle<edm::View<pat::Jet> > jets;
iEvent.getByLabel("selectedPatJets",jets);
Handle<ValueMap<float> > puJetIdMVA;
iEvent.getByLabel("fullDiscriminant",puJetMva);
Handle<ValueMap<int> > puJetIdFlag;
iEvent.getByLabel("fullId",puJetMva);
for ( unsigned int i=0; i<jets->size(); ++i ) {
const pat::Jet & patjet = jets->at(i);
float mva = (*puJetIdMVA)[jets->refAt(i)];
int idflag = (*puJetIdFlag)[jets->refAt(i)];
cout << "jet " << i << " pt " << patjet.pt() << " eta " << patjet.eta() << " PU JetID MVA " << mva;
if( PileupJetIdentifier::passJetId( idflag, PileupJetIdentifier::kLoose ) {
cout << " pass loose wp";
}
if( PileupJetIdentifier::passJetId( idflag, PileupJetIdentifier::kMedium ) {
cout << " pass medium wp";
}
if( PileupJetIdentifier::passJetId( idflag, PileupJetIdentifier::kTight ) {
cout << " pass tight wp";
}
cout << endl;
}
The ValueMap with StoredPileupJetIdentifier can be used to check the inputs to the algorithm and it allows to re-evaluate the final MVA and working
points at a later stage directly on the PAT-tuples. This can be done in the following way.
# load the PU JetID sequence
process.load("CMGTools.External.pujetidsequence_cff")
# re-evaluate PU JetID MVA
process.p = cms.Path(process.puJetMva)
# keep the PU JetID products
process.out.extend(["keep *_puJetId_*_*", # input variables
"keep *_puJetMva_*_*" # final MVAs and working point flags
])
Running the algorithm on CHS jets
A specific training for CHS jets is available. In order to use it, the above instructions can be followed, replacing the producerpuJetId with puJetIdChs and similarly puJetMva with puJetMvaChs.
Running the algorithm on reco jets
The producer can be run on collections of reco Jetsm however this feature has not been validated. The final MVA has to be evaluated on corrected jets, thus the JEC have to be fed to the producer. The producer can run both on corrected and uncorrected jets, provided that the parameters are properly set. The snippet below shows how to run on a collection of uncorrected jets. To run on corrected reco jets, set theapplyJec flag to False and the inputIsCorrected to True.
from CMGTools.External.pujetidsequence_cff import puJetId
process.recoPuJetId = puJetId.clone(
jets = cms.InputTag("ak5PFJets"),
applyJec = cms.bool(True),
inputIsCorrected = cms.bool(False),
)
process.recoPuJetMva = puJetMva.clone(
jets = cms.InputTag("ak5PFJets"),
jetids = cms.InputTag("recoPuJetId"),
applyJec = cms.bool(True),
inputIsCorrected = cms.bool(False),
)
process.recoPuJetIdSqeuence(process.recoPuJetId * process.recoPuJetMva )
Edit | Attach | Topic revision: r2 - 2017-10-05 - EricSchanet
Home 
Plugins
Sandbox for tests
Support
Alice
Atlas
CMS
LHCb
Public Webs
Sandbox Web 
Create New Topic
Index
Search
Changes
Notifications
RSS Feed
Statistics
Account 
Copyright &© 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors. or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback