Z(υυ) H(bb) Step by Step Analysis

.. NOTE:: All the scripts must be run from the base directory. Use: source /cvmfs/cms.cern.ch/slc5_amd64_gcc462/lcg/root/5.34.02-cms/bin/thisroot.sh

1. Use maketier2list.py to transfer Step 2 ntuples from Pisa to FNAL. edit the directories in the script before run.

python maketier2list.py

2. Use Skim.C or Skim_backup.C to skim the Step 2 ntuples with baseline selection. Make sure HelperNtuples.h is updated. in inputstep2.ini Section [Skim], edit tagMC, tagData, baseline, mettrigger, metfilter. in inputstep2.ini Section [Stitch], edit xxxLHECUT's enable reader.write_HelperNtuples(), disable the rest python pyhelper.py copy printout to HelperNtuples.h now run the skim

source run_Skim.sh

3. Use SkimRegression.C to skim the Step2 ntuples for BDTG regression training. Make sure HelperNtuples.h is updated. in inputstep2.ini Section [Skim], edit regression, fjregression enable reader.write_HelperNtuples() in pyhelper.py, disable the rest

python pyhelper.py copy printout to HelperNtuples.h: python pyhelper.py > HelperNtuples.h now run the skim for ak5 regression

source run_SkimRegression.sh now run the skim for filter jet regression source run_SkimRegressionFJ.sh

4. Use TrainRegression.C and TrainRegressionFJ.C to produce the BDT regression .xml files. Make sure HelperTMVA.h is updated. in inputstep2.ini Section [BDT Regression Variable], edit the variables to use in inputstep2.ini Section [BDT Regression FJ Variable], edit the variables to use enable reader.write_HelperTMVA() in pyhelper.py, disable the rest python pyhelper.py copy printout to HelperTMVA.h: python pyhelper.py > HelperTMVA.h now run the BDT regression python run_TrainRegression.py cp weights/TMVARegression_BDTG.weights.xml weights/TMVARegression_BDTG.testweights.xml cp TMVAReg.root testTMVAReg.root now run the BDT regression for FJ python run_TrainRegressionFJ.py cp weights/TMVARegressionFJ_BDTG.weights.xml weights/TMVARegressionFJ_BDTG.testweights.xml cp TMVARegFJ.root testTMVARegFJ.root

To check the regression performances run python ComparePtResolution.py python ComparePtOffset.py python CompareMass_sig.py

5. Update HelperNtuples.h to have all the correct numbers. enable skimmer.process() in skimmer.py, disable the rest python skimmer.py inputstep2.ini copy printout to inputstep2.ini Section [Process] enable skimmer.stitch() in skimmer.py, disable the rest python skimmer.py inputstep2.ini copy printout to inputstep2.ini Section [Stitch] enable reader.write_HelperNtuples() in pyhelper.py, disable the rest python pyhelper.py copy printout to HelperNtuples.h: python pyhelper.py > HelperNtuples.h

6. Use GrowTree.C to create Step 3's. python run_GrowTree.py

7. Use SkimClassification.C to produce the files for BDT classification. source run_SkimClassification.sh

8. Use TrainBDT.C to produce the BDT regression .xml files. source run_TrainBDT.sh

9. Use TrimTree.C to create Step 4's. in inputstep2.ini Section [Weight], edit the MC event weights in inputstep2.ini Section [Trigger], edit the Data trigger in inputstep2.ini Section [Selection], edit the signal and control regions enable reader.write_HelperTMVA() in pyhelper.py, disable the rest python pyhelper.py copy printout to HelperTMVA.h: python pyhelper.py > HelperTMVA.h in TrimTree.C, edit the BDT .xml files to load, then run python run_TrimTree.py stitch the Step 4's (remember to edit $DIR in the script) source run_stitch.sh

10. run for now cd macros root -l -b -q plotHistos_Znn_13TeV_BDT.C++ source results_Znn.csh

later we should do 10. Use ScaleFactorJ11.C and ScaleFactorWorkspaceJ11.C to fit scale factors. .. code:: bash set the variable indir in the scripts, reset the fitresults arrays, reset the iteration number in calc_scalefactors(...) root -l -b -q ScaleFactorJ11.C++O root -l -b -q ScaleFactorWorkspaceJ11.C++O use MaxLikelihoodFit to get the nuisances combineCards.py ZnunuHighPt_WjLF=vhbb_Znn_SF_J11_ZnunuHighPt_WjLF_8TeV.txt ZnunuHighPt_WjHF=vhbb_Znn_SF_J11_ZnunuHighPt_WjHF_8TeV.txt ZnunuHighPt_ZjLF=vhbb_Znn_SF_J11_ZnunuHighPt_ZjLF_8TeV.txt ZnunuHighPt_ZjHF=vhbb_Znn_SF_J11_ZnunuHighPt_ZjHF_8TeV.txt ZnunuHighPt_TT=vhbb_Znn_SF_J11_ZnunuHighPt_TT_8TeV.txt > vhbb_Znn_SF_J11_ZnunuHighPt_8TeV.txt combine -M MaxLikelihoodFit -m 125 --robustFit=1 --stepSize=0.05 --rMin=-5 --rMax=5 --saveNorm vhbb_Znn_SF_J11_ZnunuHighPt_8TeV.txt python printNuisances.py mlfit.root

11. Use BDTShapeJ12.C and BDTShapeWorkspaceJ12.C to get final limit and significance. set the variable indir in the scripts, adjust binning, then run root -l -b -q BDTShapeJ12.C++O root -l -b -q BDTShapeWorkspaceJ12.C++O calculate expected limit and significance combine -M Asymptotic -t -1 vhbb_Znn_J11_ZnunuHighPt_8TeV.txt combine -M ProfileLikelihood -m 125 --signif --pvalue -t -1 --toysFreq --expectSignal=1 vhbb_Znn_J11_ZnunuHighPt_8TeV.txt combine -M MaxLikelihoodFit -m 125 --robustFit=1 --stepSize=0.05 --rMin=-5 --rMax=5 --saveNorm -t -1 --toysFreq --expectSignal=1 vhbb_Znn_J11_ZnunuHighPt_8TeV.txt python diffNuisances.py -f html mlfit.root calculate observed limit and significance combine -M Asymptotic vhbb_Znn_J11_ZnunuHighPt_8TeV.txt combine -M ProfileLikelihood -m 125 --signif --pvalue vhbb_Znn_J11_ZnunuHighPt_8TeV.txt combine -M MaxLikelihoodFit -m 125 --robustFit=1 --stepSize=0.05 --rMin=-5 --rMax=5 --saveNorm vhbb_Znn_J11_ZnunuHighPt_8TeV.txt python diffNuisances.py -f html mlfit.root

Tuning

1. In directory tuneBDT, use prepare_tuneBDT.sh to create symbolic links. This step is only needed at the first time. edit the links if necessary source prepare_tuneBDT.sh

2. Use BDTShapeJ12_reload.C.diff to patch BDTShapeJ12_reload.C.

the patch was created with diff -u BDTShapeJ12_reload.C.orig BDTShapeJ12_reload.C it might only work with a specific version of BDTShapeJ12.C patch -p0 < BDTShapeJ12_reload.C.diff

3. Use run_tune.py to get commands to train and reload. edit r, n, w if necessary python run_tune.py first train, then reload

4. Use run_combine_tune.py to calculate final limit and significance. .. code:: bash edit r, n, w, logs if necessary python run_combine_tune.py

5. Use retrieve_combine_tune.py to store the previous results into TTrees. .. code:: bash edit r, w, logs if necessary it only does one channel at a time python run_combine_tune.py use hadd to combine the TTrees

6. Open the combined TTree (e.g. TMVA_ZnunuHighPt_new.root), scan for the best significance. .. code:: bash root [0] fomtree->Scan("NTrees:nEventsMin:MaxDepth:COMB_limit:COMB_signif:TMVA_kolS:TMVA_kolB:USER_psig","COMB_limit<3.6 && COMB_signif>0.68")

7. BDT binning optimization instructions pending...

8. Find yield uncertainties (only for ZbbHinv cards). .. code:: bash edit nuisances, processes, soverb if necessary python systematica_ZbbHinv.py