H->cc on top of rhalphalib

Install

Following the recipe from combine. and clone to CMSSW environment.

export SCRAM_ARCH=slc7_amd64_gcc700
cmsrel CMSSW_10_2_13
cd CMSSW_10_2_13/src
cmsenv
git clone https://github.com/cms-analysis/HiggsAnalysis-CombinedLimit.git HiggsAnalysis/CombinedLimit
cd HiggsAnalysis/CombinedLimit
cd $CMSSW_BASE/src/HiggsAnalysis/CombinedLimit
git fetch origin
git checkout v8.0.1
scramv1 b clean; scramv1 b

cd $CMSSW_BASE/src/
git clone https://github.com/cms-analysis/CombineHarvester.git CombineHarvester
scram b -j8

Then install rhalphalib

cmsenv
cd $CMSSW_BASE/src/
git clone git@github.com:andrzejnovak/rhalphalib.git
cd rhalphalib
git fetch
git checkout origin/newhcc
# Need to update some packages against the ones in CMSSW (might need a few more)
pip install uproot --user --upgrade
pip install matplotlib --user --upgrade
pip install mplhep --user
# Install rhalphalib 
pip install --user -e .

and finally clone and run higgstocharm

git clone https://github.com/andrzejnovak/higgstocharm.git
cd higgstocharm

# Must chose --data or --MC, other options get printed
python new_Hxx.py --data --unblind --year 2017 --templates n2nano/templates_nskim17_CC.root -o Test17

Fitting

Building workspace commands

bash build.sh
# text2workspace.py -P HiggsAnalysis.CombinedLimit.PhysicsModel:multiSignalModel  --PO verbose --PO 'map=.*/*hcc*:r[1,-500,500]' --PO 'map=.*/zcc:z[1,-5,5]' model_combined.txt
# text2workspace.py -P HiggsAnalysis.CombinedLimit.PhysicsModel:multiSignalModel  --PO verbose --PO 'map=.*/*hcc*:r[1,-500,500]' model_combined.txt
# text2workspace.py -P HiggsAnalysis.CombinedLimit.PhysicsModel:multiSignalModel  --PO verbose --PO 'map=.*/zcc:r[1,-5,5]' model_combined.txt

combine -M FitDiagnostics --expectSignal 1 -d model_combined.root --cminDefaultMinimizerStrategy 0 --robustFit=1 -t -1 --toysFrequentist 
combine -M FitDiagnostics --expectSignal 1 -d model_combined.root --cminDefaultMinimizerStrategy 0 --robustFit=1 --saveShapes --saveWithUncertainties -t -1 --toysFrequentist 
combine -M Significance model_combined.root --expectSignal 1  -t -1 --toysFrequentist
combineTool.py -M AsymptoticLimits -m 125 -d model_combined.root --there --expectSignal 1 -t -1 --toysFrequentist
python ../HiggsAnalysis/CombinedLimit/test/diffNuisances.py tempModel/fitDiagnostics.root 


python ../plot.py --data 
python ../plotTF.py

# python ../plot.py --MC --year 2017 -o plots_MC_t1

OTHER IMPORTANT COMMANDS

combineTool.py -M FitDiagnostics -m 125 -d model_combined.root --there --cminDefaultMinimizerStrategy 0 -t -1 --expectSignal 1
combineTool.py -M AsymptoticLimits -m 125 -d model_combined.root --there -t -1 --expectSignal 1 --rMin=-50 --rMax=50

combine -M FitDiagnostics -t -1 --expectSignal 0 -d tempModel_combined.root --rMin=-5 --rMax=10  --cminDefaultMinimizerStrategy 0 --robustFit=1

Running Impacts

# Baseline
combineTool.py -M Impacts -d model_combined.root -m 125 --doInitialFit --robustFit 1 --setParameterRanges r=-1,5 --cminDefaultMinimizerStrategy 0 --X-rtd FITTER_DYN_STEP --expectSignal 1 -t -1 --toysFrequentist 
# Condor
combineTool.py -M Impacts -d model_combined.root -m 125 --doFits --robustFit 1 --allPars --setParameterRanges r=-1,5  -t -1 --toysFrequentist --expectSignal 1 --cminDefaultMinimizerStrategy 0 --X-rtd MINIMIZER_analytic --job-mode condor --sub-opts='+JobFlavour = "workday"' --task-name ggHccFit --exclude 'rgx{qcdparams*}'
# Collect
combineTool.py -M Impacts -d model_combined.root -m 125 --allPars -o impacts.json
plotImpacts.py -i impacts.json -o plots/impacts_out --blind

Running likelihood scan

combineTool.py -M MultiDimFit -d model_combined.root --cminDefaultMinimizerStrategy 0 --expectSignal 1 --robustFit 1 --algo grid --points 40 --setParameterRanges r=-20,20 -m 125 -t -1 --toysFrequentist

plot1DScan.py higgsCombine.Test.MultiDimFit.mH125.root -o plots/LScan_data_Zexp1 --y-max 30 --y-cut 30

Running tests

Requirements

Standalone model creation requires:

• Python 2.7+ or 3.6+
• numpy >= 1.14

RooFit+combine rendering requires:

• ROOT < 6.18 (i.e. LCG96 is too recent, CMSSW 8 combine cannot handle it. LCG95a is fine)

Use in combine requires, well, combine. The CMSSW 10 (CC7) recipe satisfies the requirements, however the CMSSW 8 recipe has a too old version of numpy.

There is a python 3 compatible standalone fork of combine available. It is also possible to render the model folder using the quickstart recipe, and then move the folder or switch environments to a CMSSW+combine environment and proceed from there.

################################# ##############################

Shit that worked

# For Zcc, all systs
combine -M FitDiagnostics --expectSignal 1 -d tempModel_combined.root --rMin=-1 --rMax=3  --cminDefaultMinimizerStrategy 0 --robustFit=1  -t -1 --toysFrequentist