TauSF

a repository for fitting the tau ID SFs using ZTT events

Setup

setup CMSSW:

 cmsrel CMSSW_10_2_13
 cd CMSSW_10_2_13/src
 cmsenv

Clone combine and combine harvester:

 git clone -b 102x git@github.com:cms-analysis/CombineHarvester.git CombineHarvester
 git clone https://github.com/cms-analysis/HiggsAnalysis-CombinedLimit.git HiggsAnalysis/CombinedLimit
 cd HiggsAnalysis/CombinedLimit
 git fetch origin
 git checkout v8.2.0
 cd -

Clone tauSF repository:

git clone git@github.com:danielwinterbottom/TauSF.git CombineHarvester/TauSF

compile:

scram b clean
scram b -j8

Copy datacards to shapes directory

Add instructions to retrieve these from a repository

....

Converting datacards

If you are starting from 2D histograms in bins of pT vs m_vis these must first be split into 1D histograms in bins of m_vis Run this script to split the datacards:

python scripts/convertDatacards.py -f shapes/ztt.datacard.pt_2_vs_m_vis.mt.2017.root

Produce txt datacards

Define your output folder:

output_dir="tauSF_output"

The various options are set in the config/harvestDatacards.yml config file

Run the following script to produce the txt datacards

python scripts/harvestDatacards.py -o ${output_dir}

Create workspace

combineTool.py -M T2W -i outputs/${output_dir}/cmb/ -o ws.root --X-allow-no-signal

Run fits

Run fits to determine proper values and errors of all POIs

When using DM-binned fits define POIs with:

pois="rate_tauSF_DM0_pT20to25_2016_preVFP,rate_tauSF_DM0_pT25to30_2016_preVFP,rate_tauSF_DM0_pT30to35_2016_preVFP,rate_tauSF_DM0_pT35to40_2016_preVFP,rate_tauSF_DM0_pT40to50_2016_preVFP,rate_tauSF_DM0_pT50to60_2016_preVFP,rate_tauSF_DM0_pT60to80_2016_preVFP,rate_tauSF_DM0_pT80to100_2016_preVFP,rate_tauSF_DM0_pT100to200_2016_preVFP,rate_tauSF_DM1_pT20to25_2016_preVFP,rate_tauSF_DM1_pT25to30_2016_preVFP,rate_tauSF_DM1_pT30to35_2016_preVFP,rate_tauSF_DM1_pT35to40_2016_preVFP,rate_tauSF_DM1_pT40to50_2016_preVFP,rate_tauSF_DM1_pT50to60_2016_preVFP,rate_tauSF_DM1_pT60to80_2016_preVFP,rate_tauSF_DM1_pT80to100_2016_preVFP,rate_tauSF_DM1_pT100to200_2016_preVFP,rate_tauSF_DM10_pT20to25_2016_preVFP,rate_tauSF_DM10_pT25to30_2016_preVFP,rate_tauSF_DM10_pT30to35_2016_preVFP,rate_tauSF_DM10_pT35to40_2016_preVFP,rate_tauSF_DM10_pT40to50_2016_preVFP,rate_tauSF_DM10_pT50to60_2016_preVFP,rate_tauSF_DM10_pT60to80_2016_preVFP,rate_tauSF_DM10_pT80to100_2016_preVFP,rate_tauSF_DM10_pT100to200_2016_preVFP,rate_tauSF_DM11_pT20to25_2016_preVFP,rate_tauSF_DM11_pT25to30_2016_preVFP,rate_tauSF_DM11_pT30to35_2016_preVFP,rate_tauSF_DM11_pT35to40_2016_preVFP,rate_tauSF_DM11_pT40to50_2016_preVFP,rate_tauSF_DM11_pT50to60_2016_preVFP,rate_tauSF_DM11_pT60to80_2016_preVFP,rate_tauSF_DM11_pT80to100_2016_preVFP,rate_tauSF_DM11_pT100to200_2016_preVFP,rate_tauSF_DM0_pT20to25_2016_postVFP,rate_tauSF_DM0_pT25to30_2016_postVFP,rate_tauSF_DM0_pT30to35_2016_postVFP,rate_tauSF_DM0_pT35to40_2016_postVFP,rate_tauSF_DM0_pT40to50_2016_postVFP,rate_tauSF_DM0_pT50to60_2016_postVFP,rate_tauSF_DM0_pT60to80_2016_postVFP,rate_tauSF_DM0_pT80to100_2016_postVFP,rate_tauSF_DM0_pT100to200_2016_postVFP,rate_tauSF_DM1_pT20to25_2016_postVFP,rate_tauSF_DM1_pT25to30_2016_postVFP,rate_tauSF_DM1_pT30to35_2016_postVFP,rate_tauSF_DM1_pT35to40_2016_postVFP,rate_tauSF_DM1_pT40to50_2016_postVFP,rate_tauSF_DM1_pT50to60_2016_postVFP,rate_tauSF_DM1_pT60to80_2016_postVFP,rate_tauSF_DM1_pT80to100_2016_postVFP,rate_tauSF_DM1_pT100to200_2016_postVFP,rate_tauSF_DM10_pT20to25_2016_postVFP,rate_tauSF_DM10_pT25to30_2016_postVFP,rate_tauSF_DM10_pT30to35_2016_postVFP,rate_tauSF_DM10_pT35to40_2016_postVFP,rate_tauSF_DM10_pT40to50_2016_postVFP,rate_tauSF_DM10_pT50to60_2016_postVFP,rate_tauSF_DM10_pT60to80_2016_postVFP,rate_tauSF_DM10_pT80to100_2016_postVFP,rate_tauSF_DM10_pT100to200_2016_postVFP,rate_tauSF_DM11_pT20to25_2016_postVFP,rate_tauSF_DM11_pT25to30_2016_postVFP,rate_tauSF_DM11_pT30to35_2016_postVFP,rate_tauSF_DM11_pT35to40_2016_postVFP,rate_tauSF_DM11_pT40to50_2016_postVFP,rate_tauSF_DM11_pT50to60_2016_postVFP,rate_tauSF_DM11_pT60to80_2016_postVFP,rate_tauSF_DM11_pT80to100_2016_postVFP,rate_tauSF_DM11_pT100to200_2016_postVFP,rate_tauSF_DM0_pT20to25_2017,rate_tauSF_DM0_pT25to30_2017,rate_tauSF_DM0_pT30to35_2017,rate_tauSF_DM0_pT35to40_2017,rate_tauSF_DM0_pT40to50_2017,rate_tauSF_DM0_pT50to60_2017,rate_tauSF_DM0_pT60to80_2017,rate_tauSF_DM0_pT80to100_2017,rate_tauSF_DM0_pT100to200_2017,rate_tauSF_DM1_pT20to25_2017,rate_tauSF_DM1_pT25to30_2017,rate_tauSF_DM1_pT30to35_2017,rate_tauSF_DM1_pT35to40_2017,rate_tauSF_DM1_pT40to50_2017,rate_tauSF_DM1_pT50to60_2017,rate_tauSF_DM1_pT60to80_2017,rate_tauSF_DM1_pT80to100_2017,rate_tauSF_DM1_pT100to200_2017,rate_tauSF_DM10_pT20to25_2017,rate_tauSF_DM10_pT25to30_2017,rate_tauSF_DM10_pT30to35_2017,rate_tauSF_DM10_pT35to40_2017,rate_tauSF_DM10_pT40to50_2017,rate_tauSF_DM10_pT50to60_2017,rate_tauSF_DM10_pT60to80_2017,rate_tauSF_DM10_pT80to100_2017,rate_tauSF_DM10_pT100to200_2017,rate_tauSF_DM11_pT20to25_2017,rate_tauSF_DM11_pT25to30_2017,rate_tauSF_DM11_pT30to35_2017,rate_tauSF_DM11_pT35to40_2017,rate_tauSF_DM11_pT40to50_2017,rate_tauSF_DM11_pT50to60_2017,rate_tauSF_DM11_pT60to80_2017,rate_tauSF_DM11_pT80to100_2017,rate_tauSF_DM11_pT100to200_2017,rate_tauSF_DM0_pT20to25_2018,rate_tauSF_DM0_pT25to30_2018,rate_tauSF_DM0_pT30to35_2018,rate_tauSF_DM0_pT35to40_2018,rate_tauSF_DM0_pT40to50_2018,rate_tauSF_DM0_pT50to60_2018,rate_tauSF_DM0_pT60to80_2018,rate_tauSF_DM0_pT80to100_2018,rate_tauSF_DM0_pT100to200_2018,rate_tauSF_DM1_pT20to25_2018,rate_tauSF_DM1_pT25to30_2018,rate_tauSF_DM1_pT30to35_2018,rate_tauSF_DM1_pT35to40_2018,rate_tauSF_DM1_pT40to50_2018,rate_tauSF_DM1_pT50to60_2018,rate_tauSF_DM1_pT60to80_2018,rate_tauSF_DM1_pT80to100_2018,rate_tauSF_DM1_pT100to200_2018,rate_tauSF_DM10_pT20to25_2018,rate_tauSF_DM10_pT25to30_2018,rate_tauSF_DM10_pT30to35_2018,rate_tauSF_DM10_pT35to40_2018,rate_tauSF_DM10_pT40to50_2018,rate_tauSF_DM10_pT50to60_2018,rate_tauSF_DM10_pT60to80_2018,rate_tauSF_DM10_pT80to100_2018,rate_tauSF_DM10_pT100to200_2018,rate_tauSF_DM11_pT20to25_2018,rate_tauSF_DM11_pT25to30_2018,rate_tauSF_DM11_pT30to35_2018,rate_tauSF_DM11_pT35to40_2018,rate_tauSF_DM11_pT40to50_2018,rate_tauSF_DM11_pT50to60_2018,rate_tauSF_DM11_pT60to80_2018,rate_tauSF_DM11_pT80to100_2018,rate_tauSF_DM11_pT100to200_2018"

For for DM-inclusive fits define POIs with:

pois="rate_tauSF_DMinclusive_pT20to25_2016_preVFP,rate_tauSF_DMinclusive_pT25to30_2016_preVFP,rate_tauSF_DMinclusive_pT30to35_2016_preVFP,rate_tauSF_DMinclusive_pT35to40_2016_preVFP,rate_tauSF_DMinclusive_pT40to50_2016_preVFP,rate_tauSF_DMinclusive_pT50to60_2016_preVFP,rate_tauSF_DMinclusive_pT60to80_2016_preVFP,rate_tauSF_DMinclusive_pT80to100_2016_preVFP,rate_tauSF_DMinclusive_pT100to200_2016_preVFP,rate_tauSF_DMinclusive_pT20to25_2016_postVFP,rate_tauSF_DMinclusive_pT25to30_2016_postVFP,rate_tauSF_DMinclusive_pT30to35_2016_postVFP,rate_tauSF_DMinclusive_pT35to40_2016_postVFP,rate_tauSF_DMinclusive_pT40to50_2016_postVFP,rate_tauSF_DMinclusive_pT50to60_2016_postVFP,rate_tauSF_DMinclusive_pT60to80_2016_postVFP,rate_tauSF_DMinclusive_pT80to100_2016_postVFP,rate_tauSF_DMinclusive_pT100to200_2016_postVFP,rate_tauSF_DMinclusive_pT20to25_2017,rate_tauSF_DMinclusive_pT25to30_2017,rate_tauSF_DMinclusive_pT30to35_2017,rate_tauSF_DMinclusive_pT35to40_2017,rate_tauSF_DMinclusive_pT40to50_2017,rate_tauSF_DMinclusive_pT50to60_2017,rate_tauSF_DMinclusive_pT60to80_2017,rate_tauSF_DMinclusive_pT80to100_2017,rate_tauSF_DMinclusive_pT100to200_2017,rate_tauSF_DMinclusive_pT20to25_2018,rate_tauSF_DMinclusive_pT25to30_2018,rate_tauSF_DMinclusive_pT30to35_2018,rate_tauSF_DMinclusive_pT35to40_2018,rate_tauSF_DMinclusive_pT40to50_2018,rate_tauSF_DMinclusive_pT50to60_2018,rate_tauSF_DMinclusive_pT60to80_2018,rate_tauSF_DMinclusive_pT80to100_2018,rate_tauSF_DMinclusive_pT100to200_2018"

Run the fit with all systematics floating using:

combineTool.py -m 125 -M MultiDimFit --redefineSignalPOIs "${pois}" --saveWorkspace --X-rtd MINIMIZER_analytic --expectSignal 0 --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 --algo singles --cl=0.68 --there -n ".ztt.bestfit.singles" -d outputs/${output_dir}/cmb/ws.root

For decomposition of the uncertainties into groups that allow uncertainties to be correlated by era and by bins we need to run the following additional fits:

First fun a fit loading the snapshot but not freezing any uncertainties

combineTool.py -m 125 -M MultiDimFit --redefineSignalPOIs "${pois}" --saveWorkspace --X-rtd MINIMIZER_analytic --expectSignal 0 --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 --algo singles --cl=0.68 --there -d outputs/${output_dir}/cmb/higgsCombine.ztt.bestfit.singles.MultiDimFit.mH125.root -n ".ztt.bestfit.singles.postfit" --snapshotName MultiDimFit

Then run with uncertainties correlated by era and bins frozen:

combineTool.py -m 125 -M MultiDimFit --redefineSignalPOIs "${pois}" --saveWorkspace --X-rtd MINIMIZER_analytic --expectSignal 0 --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 --algo singles --cl=0.68 --there -d outputs/${output_dir}/cmb/higgsCombine.ztt.bestfit.singles.MultiDimFit.mH125.root -n ".ztt.bestfit.singles.postfit.freeze_byErasAndBins" --snapshotName MultiDimFit  --snapshotName MultiDimFit --freezeNuisanceGroups byErasAndBins

Now run with uncertainties correlated by pT bins and eras frozen as well

combineTool.py -m 125 -M MultiDimFit --redefineSignalPOIs "${pois}" --saveWorkspace --X-rtd MINIMIZER_analytic --expectSignal 0 --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 --algo singles --cl=0.68 --there -d outputs/${output_dir}/cmb/higgsCombine.ztt.bestfit.singles.MultiDimFit.mH125.root -n ".ztt.bestfit.singles.postfit.freeze_byErasAndBins_byBins" --snapshotName MultiDimFit  --snapshotName MultiDimFit --freezeNuisanceGroups byErasAndBins,byBins

For dm-binned SF only we also need to run this fit:

combineTool.py -m 125 -M MultiDimFit --redefineSignalPOIs "${pois}" --saveWorkspace --X-rtd MINIMIZER_analytic --expectSignal 0 --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 --algo singles --cl=0.68 --there -d outputs/${output_dir}/cmb/higgsCombine.ztt.bestfit.singles.MultiDimFit.mH125.root -n ".ztt.bestfit.singles.postfit.freeze_byErasAndBins_byBins_byDM" --snapshotName MultiDimFit  --snapshotName MultiDimFit --freezeNuisanceGroups byErasAndBins,byBins,byDM0,byDM1,byDM10,byDM11

Old: Once the fits are run it is also possible to make a summary plot of the uncertainty magnitudes using:

scripts/compareSFUncerts.py

Note the names of the graphs are currently hardcoded in this script

The following instructions will prepare the graphs and then decouple the uncertainties storing the results to root files and making plots

for pT-dependent fits:

mkdir tau_sf_medium

for x in "" .freeze_byErasAndBins .freeze_byErasAndBins_byBins ; do python scripts/makeSFGraphs.py  -f outputs/${output_dir}/cmb/higgsCombine.ztt.bestfit.singles.postfit${x}.MultiDimFit.mH125.root; done

dir=outputs/${output_dir}/cmb/

python scripts/decoupleUncerts.py -f1 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.MultiDimFit.mH125.TGraphAsymmErrors.root -f2 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.freeze_byErasAndBins.MultiDimFit.mH125.TGraphAsymmErrors.root -f3 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.freeze_byErasAndBins_byBins.MultiDimFit.mH125.TGraphAsymmErrors.root -o tau_sf_medium

for dm-binned fits:

mkdir tau_sf_medium

for x in "" .freeze_byErasAndBins .freeze_byErasAndBins_byBins .freeze_byErasAndBins_byBins_byDM; do python scripts/makeSFGraphs.py  -f outputs/${output_dir}/cmb/higgsCombine.ztt.bestfit.singles.postfit${x}.MultiDimFit.mH125.root --dm-bins; done

dir=outputs/${output_dir}/cmb/

python scripts/decoupleUncerts.py -f1 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.MultiDimFit.mH125.TGraphAsymmErrors.root -f2 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.freeze_byErasAndBins.MultiDimFit.mH125.TGraphAsymmErrors.root -f3 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.freeze_byErasAndBins_byBins.MultiDimFit.mH125.TGraphAsymmErrors.root  -f4 ${dir}/higgsCombine.ztt.bestfit.singles.postfit.freeze_byErasAndBins_byBins_byDM.MultiDimFit.mH125.TGraphAsymmErrors.root --dm-bins -o tau_sf_medium

Run postfit plots

First need to run a fit and store the fit result. We use the robustHesse option to make sure the covariance matrix is accurate

combineTool.py -m 125 -M MultiDimFit --redefineSignalPOIs "${pois}" --saveFitResult --X-rtd MINIMIZER_analytic --expectSignal 0 --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 --algo none --there -n ".ztt.bestfit.singles.robustHesse" -d outputs/${output_dir}/cmb/ws.root --robustHesse=1

Make the histograms containing the postfit shapes. Note this script only computes the uncertainties for the TotalBkg histogram not for the individual processes to save time (note ZTT is included as background in the current setup)

python python/PostFitShapesCombEras.py -f outputs/${output_dir}/cmb/multidimfit.ztt.bestfit.singles.robustHesse.root:fit_mdf -w outputs/${output_dir}/cmb/ws.root -d outputs/${output_dir}/cmb/combined.txt.cmb --output shapes_postfit.root

You can optionally use the option "-b" to only specify one bin at a time allowing the shapes to be run parallel e.g "-b ztt_mt_1_2018"

Make the plots by running the script "scripts/postFitPlots.py"

e.g to make the plots for all eras and pT-dependent bins:

for b in 1 2 3 4 5 6 7 8 9; do for era in 2016_preVFP 2016_postVFP 2017 2018; do python scripts/postFitPlots.py --file shapes_postfit.root --file_dir ztt_mt_${b}_${era} --ratio --ratio_range 0.8,1.2; done; done

Run impacts (approximate)

Do fits:

combineTool.py -M Impacts  --doFits --robustFit=1 --approx robust --redefineSignalPOIs rate_tauSF_DM1_pT40to50_2018 -d outputs/tauSF_output_DMbinned/cmb/ws.root  -n ".ztt.impacts"  --X-rtd MINIMIZER_analytic --cminDefaultMinimizerStrategy 0 --cminDefaultMinimizerTolerance 0.1 -m 125

Collect results:

combineTool.py -M Impacts --approx robust --redefineSignalPOIs rate_tauSF_DM1_pT40to50_2018 -d outputs/tauSF_output_DMbinned/cmb/ws.root  -n ".ztt.impacts" -o impacts.json -m 125

Make plot:

plotImpacts.py -i impacts.json -o impacts

Make TGraphAsymmErrors for fitted results. For dm-binned SF use option --dm-bins - this will also perform fits of the SFs vs pT

Produce the graphs:

python scripts/makeSFGraphs.py -f outputs/tauSF_Feb08_DMinclusive_tightVsE_mTLt30/cmb/higgsCombine.ztt.bestfit.singles.MultiDimFit.mH125.root

for pT-dependent SF you can also make a plot of the graphs after running the previous script using:

python scripts/plotInclusiveSF.py -f outputs/tauSF_Feb08_DMinclusive_tightVsE_mTLt30/cmb/higgsCombine.ztt.bestfit.singles.MultiDimFit.mH125.TGraphAsymmErrors.root