git clone git@github.com:cbernet/recnn.git 
cd recnn
ln -s /data/conda/recnn/data
ln -s /data/conda/recnn/data_gilles_louppe
ln -s /data/conda/recnn/models

Pour executer certain script de plotting il vous faudra aussi recuperer le package cpyroot:

cd ..
git clone git@github.com:cbernet/cpyroot.git
cd cpyroot
source init.sh
cd ../recnn

Original paper

The signal and background samples come from the CMS simulation, and are stored into ROOT files:

/data2/conda/recnn/data/rootfiles/RawSignal_21Sept.root
/data2/conda/recnn/data/rootfiles/RawBackground_17july.root  

All Jets in the Background rootfile are to be considered as background (not hadronic taus) whereas jets in the Signal rootfile are to be considered signal only if a gen-level Tau is found very close to the jet:

dRs[0] != 0. && dRs[0] < 0.3

Original samples :

• signal :

  /SUSYGluGluToHToTauTau_M-*_TuneCUETP8M1_13TeV-pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /SUSYGluGluToBBHToTauTau_M-*_TuneCUETP8M1_13TeV-pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  with * in [80,90,100,110,120,130,140,160,180,200,250,350,400,450,500,600,700,800,900,1000,1200,1400,1600,1800,2000,2300,2600,2900,3200]

• background :

  /QCD_Pt_15to30_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /QCD_Pt_30to50_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /QCD_Pt_50to80_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /QCD_Pt_80to120_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /QCD_Pt_80to120_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6_ext2-v1/MINIAODSIM
  /QCD_Pt_120to170_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /QCD_Pt_120to170_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6_ext1-v1/MINIAODSIM
  /QCD_Pt_170to300_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6-v1/MINIAODSIM
  /QCD_Pt_170to300_TuneCUETP8M1_13TeV_pythia8/RunIISummer16MiniAODv2-PUMoriond17_80X_mcRun2_asymptotic_2016_TrancheIV_v6_ext1-v1/MINIAODSIM

These files contain a TTree called tree with the following branches:

All variables names are the names of the object's method.

• Jet[1][9] : any reconstructed jet with pt>20. GeV

  • variables in order : [px,py,pz,energy,pdgId,charge,numberOfDaughters,hadronFlavour,partonFlavour]

• ptcs[200][8] : the 200 first particles of the Jet in decreasing order of pt

  • variables in order : [px,py,pz,energy,pdgId,charge,dxy,dz]

• GenJet[1][7] : closest found gen-level Jet

  • variables in order : [px,py,pz,energy,pdgId,charge,numberOfDaughters]

• genptcs[200][8] : the 200 first particles of the gen-level Jet in decreasing order of pt

  • variables in order : [px,py,pz,energy,pdgId,charge,dxy,dz]

• RawJet[1][6] : the PF jet before jet energy correction (in CMSSW : jet.uncorrectedJet(0))

  • variables in order : [px,py,pz,energy,pdgId,charge]

• Tau[1][6] : closest reconstructed Tau

  • variables in order : [px,py,pz,energy,pdgId,charge]

• GenTau[1][7] : closest reconstructed gen-level Tau

  • variables in order : [px,py,pz,energy,pdgId,charge,decayMode]

• dRs[5] : all potentially interesting dRs (when it is not specified, the words "the closest" means the closest to the reconstructed jet )

  • dRs[0] = dR between the Jet and the closest gen-level Tau
  • dRs[1] = dR between the closest gen-level Tau and its closest reconstructed Tau
  • dRs[2] = dR between the gen-level Jet and the closest gen-level Tau
  • dRs[3] = dR between the Jet and its closest reconstructed Tau
  • dRs[4] = dR between the Jet and the closest gen-level Jet

• standardID[6] : needed variables for standard ID algorithm

  • standardID[0] = tauID('byIsolationMVArun2v1DBoldDMwLTraw')
  • standardID[1] = tauID('decayModeFinding')
  • standardID[2] = tau.leadChargedHadrCand().dz()
  • standardID[3] = tauID('againstElectronVLooseMVA6')
  • standardID[4] = tauID('againstMuonLoose3')
  • standardID[5] = tau.decayMode()

• event[3] : event identification ( = serial numbers of the collision in which the jet was recorded)

  • event[0] = eventId
  • event[1] = lumi (lumisection)
  • event[2] = run

Those three steps can be done in a single cfg file that can be used like :

nohup ipython Hadronic_taus_cfg.py <path to the working directory that will be created> > Hadronic_taus_cfg.out &

The rootfiles with the input and the test results are in /rawBackground.root and /rawSignal.root for background (QCD jets) and signal (hadronic taus) respectively. We advise to put on a disk with a lot of space.

python ROC.py <path to working directory>

The produced ROC curves will be in /ROCs.root .