Electron ID for H->VV in 2012

Code to evaluate the electron ID MVA

This is a quick and dirty recipe to evaluate the MVA given inputs. It is independent on CMSSW. A cleaner implementation will be done in CMSSW.

To use it:
  • non-triggering electrons:
  fMVA->Initialize("BDTSimpleCat",
                             "elebdtweights/DanieleMVA_BDTCat_BDTG_SiDanV2.weights.xml",
                             ElectronIDMVAHZZ::kBDTSiDanV2);
  • triggering electrons:
  fMVA->Initialize("BDTSimpleCat",
                             "elebdtweights/DanieleMVA_DenomHWW_BDTCat_BDTG_SiDanV2.weights.xml",
                             ElectronIDMVAHZZ::kBDTHWWSiDanV2);

and then evaluate that with fMVA->MVAValue(...) providing the list of inputs. The weight files are also in the same CVS are.

Code to evaluate the electron PF-isolation with optimize vetoes

This code uses pf-candidates and produces association map electron-isolation. It may also compute in 'directional' isolation fashion (default is NOT directional). The python for the producer is in Emanuele's UserCode area

  • This uses as input the PF-collection after charged hadron subtraction from pileup (aka CHS or pfNoPileup), so you need to run that module first:

from CommonTools.ParticleFlow.pfNoPileUp_cff import *
pfPileUp.PFCandidates = "particleFlow"
pfNoPileUp.bottomCollection = "particleFlow"
pfPUSequence = cms.Sequence( pfPileUp * pfNoPileUp )

Effective Areas calculated on data

The following effective areas are evaluated on Z->ee events on the full 2011 data for PF isolation with cone ΔR=0.4 or ΔR=0.3. We evaluate them separately for the neutral hadron and photons in order to allow people to use them separately.

  • Isolation cone ΔR=0.4:

bin EA neutral had. EA γs
abs(η)<1.0 Aeff(NH) = 0.045 ± 0.001 Aeff(γ) = 0.14 ± 0.003
abs(η)>1.0 && abs(η)<1.479 Aeff(NH) = 0.062 ± 0.002 Aeff(γ) = 0.13 ± 0.005
abs(η)>1.479 && abs(η)<2.0 Aeff(NH) = 0.061 ± 0.002 Aeff(γ) = 0.080 ± 0.002
abs(η)>2.0 && abs(η)<2.2 Aeff(NH) = 0.041 ± 0.004 Aeff(γ) = 0.13 ± 0.004
abs(η)>2.2 && abs(η)<2.3 Aeff(NH) = 0.050 ± 0.006 Aeff(γ) = 0.14 ± 0.007
abs(η)>2.3 && abs(η)<2.4 Aeff(NH) = 0.051 ± 0.007 Aeff(γ) = 0.16 ± 0.008
abs(η)>2.4 Aeff(NH) = 0.11 ± 0.009 Aeff(γ) = 0.18 ± 0.009

  • plots for the effective areas calibrations here
  • plots for the combined average isolation sums after pileup-subtraction here

  • Isolation cone ΔR=0.3:

bin EA neutral had. EA γs
abs(η)<1.0 Aeff(NH) = 0.024 ± 0.001 Aeff(γ) = 0.080 ± 0.002
abs(η)>1.0 && abs(η)<1.479 Aeff(NH) = 0.034 ± 0.001 Aeff(γ) = 0.085 ± 0.005
abs(η)>1.479 && abs(η)<2.0 Aeff(NH) = 0.034 ± 0.002 Aeff(γ) = 0.048 ± 0.002
abs(η)>2.0 && abs(η)<2.2 Aeff(NH) = 0.013 ± 0.002 Aeff(γ) = 0.084 ± 0.004
abs(η)>2.2 && abs(η)<2.3 Aeff(NH) = 0.017 ± 0.004 Aeff(γ) = 0.085 ± 0.006
abs(η)>2.3 && abs(η)<2.4 Aeff(NH) = 0.014 ± 0.005 Aeff(γ) = 0.090 ± 0.006
abs(η)>2.4 Aeff(NH) = 0.025 ± 0.005 Aeff(γ) = 0.11 ± 0.007

* correcting PF isolation for effective areas

    • First step is calculating rho. The python module to do this in CMSSW. N.B The calculation is different with respect the one used for jet corrections because it arrives to |eta|=2.5 to be in the tracker volume to be consistent with electron acceptance, and to avoid big fluctuations.

from RecoJets.JetProducers.kt4PFJets_cfi import *
kt6PFJetsForIsolation = kt4PFJets.clone( rParam = 0.6, doRhoFastjet = True )
kt6PFJetsForIsolation.Rho_EtaMax = cms.double(2.5)

    • second step is to calculate the corrected isolation. This is simply:
      isocorr = chargediso + max(PFIso(&gamma;) - rho * Aeff(&gamma), 0.) +  max(PFIso(NH) - rho * Aeff(NH), 0.) 

Optimized working points

Cuts are given for the ID BDT / relative PF isolation.

Non triggering electrons MVA

For the optimization the samples used are:
  • signal electrons: Z→ee Monte Carlo with mc match applied.
  • fake electrons: data (W+1 jet fakeable sample, odd events: even events used for training of BDT)

  • 5<pT<10 GeV
WP /η/<0.8 0.8</η/<1.479 1.479</η/<2.5
WP 70 0.307 / 0.406 0.304 / 0.454 0.147 / 0.658
WP 80 0.100 / 0.671 0.062 / 0.655 -0.158 / 0.757
WP 85 -0.081 / 0.842 0.001 / 0.861 -0.212 / 1.064
WP 90 -0.455 / 1.147 -0.300 / 1.124 -0.537 / 1.060
WP 95 -0.286 / 48.948 -0.568 / 1.361 -0.706 / 1.400

  • pT>10 GeV
WP /η/<0.8 0.8</η/<1.479 1.479</η/<2.5
WP 70 0.975 / 0.173 0.966 / 0.195 0.930 / 0.184
WP 80 0.971 / 0.625 0.903 / 0.205 0.910 / 0.330
WP 85 0.944 / 0.306 0.708 / 0.205 0.897 / 0.522
WP 90 0.877 / 0.426 0.811 / 0.481 0.707 / 0.390
WP 95 0.634 / 0.567 0.719 / 0.909 0.593 / 0.665

Triggering electrons MVA

For the optimization the samples used are:

  • signal electrons: Z→ee Monte Carlo with mc match applied.
  • fake electrons: data (fake electron QCD triggers, odd events: even events are used for BDT training )

The H→WW fakeable object preselection is applied to both signal and background probes in the optimization.

  • the working point !WP HWW is not an optimization. Isolation and ID cuts are tuned to give the same efficiency bin-by-bin with respect the WP used for 2011 analysis. The effect on fake rate, estimated on data, is evaluated and an improvement of about 20-30% is seen. Moreover the dependency of the fake rate on pileup is small, even if the ρ corrections are applied. Plots are in this location

  • 10<pT<20 GeV
WP /η/<0.8 0.8</η/<1.479 1.479</η/<2.5
WP 70 0.390 / 0.144 0.301 / 0.129 0.574 / 0.191
WP 80 -0.013 / 0.167 0.308 / 0.213 0.397 / 0.233
WP 85 0.082 / 0.220 0.163 / 0.228 0.272 / 0.248
WP 90 -0.004 / 0.335 -0.082 / 0.291 -0.028 / 0.288
WP 95 -0.288 / 0.408 0.039 / 2.375 -0.017 / 0.463
WP HWW 0.294 / 0.158 0.730 / 0.158 0.802 / 0.108

  • pT>20 GeV:
WP /η/<0.8 0.8</η/<1.479 1.479</η/<2.5
WP 70 0.977 / 0.093 0.956 / 0.095 0.966 / 0.171
WP 80 0.913 / 0.105 0.964 / 0.178 0.899 / 0.150
WP 85 0.929 / 0.135 0.931 / 0.159 0.805 / 0.155
WP 90 0.877 / 0.177 0.794 / 0.180 0.846 / 0.244
WP 95 0.858 / 0.253 0.425 / 0.225 0.759 / 0.308
WP HWW 0.950 / 0.154 0.970 / 0.154 0.950 / 0.100

-- EmanueleDiMarco - 01-Mar-2012

Edit | Attach | Watch | Print version | History: r19 | r17 < r16 < r15 < r14 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r15 - 2012-04-02 - EmanueleDiMarco
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback