L1TauObjectsOptimization

Set of tools to evaluate tau trigger performance on T&P

Foreword

This package is based on the developments done by Olivier Davignon, Luca Cadamuro, Jean-Baptiste Sauvan, and Jona Motta. The original work done by the first three can be found mainly at these two links: NTuples production, Objects calibration

This forlder is an attempt to put together all of the developments done on top of these two folders in one single repositoty and one tool: ONE TOOL TO OPTIMIZE THEM ALL - CIT.

Install instructions

These instalation instructions work only for CMSSW_11_0_2. To use the updated CMSSW_12_0_2 version for the TauTagAndProbe package and the production of the NTuples refer to these instructions. On the other hand, still use the following instructions to do the optimization of the objects (even if NTuples are produced with CMSSW_12_0_2).

cmsrel CMSSW_11_0_2
cd CMSSW_11_0_2/src
cmsenv
git cms-init
git remote add cms-l1t-offline git@github.com:cms-l1t-offline/cmssw.git
git fetch cms-l1t-offline l1t-integration-CMSSW_11_0_2
git cms-merge-topic -u cms-l1t-offline:l1t-integration-v104.5
git cms-addpkg L1Trigger/L1TCommon
git cms-addpkg L1Trigger/L1TMuon
git clone https://github.com/cms-l1t-offline/L1Trigger-L1TMuon.git L1Trigger/L1TMuon/data
git cms-addpkg L1Trigger/L1TCalorimeter
git clone https://github.com/cms-l1t-offline/L1Trigger-L1TCalorimeter.git L1Trigger/L1TCalorimeter/data

mkdir HiggsAnalysis
cd HiggsAnalysis
git clone git@github.com:bendavid/GBRLikelihood.git
# modify the first line of `HiggsAnalysis/GBRLikelihood/BuildFile.xml` to have `-std=c++17`

git cms-checkdeps -A -a

scram b -j 10

git clone https://github.com/jonamotta/TauObjectsOptimization # package for the full optimization

Tool utilization

Production of the input objects

To produce the input objects use the TagAndProbe or TagAndProbeInegrated packages.

Merging, matching, and compression

Enter MergeTrees and run make clean ; make.

To merge the files first create/modify the needed .config file inside the MergeTrees/run directory. There the MINIAOD file has to be specified as primary and the RAW one as secondary (always use absolute paths). Check that the files really contain the TTrees that the executable will look for. Then jus run:

./merge.exe run_<year>/<optimization_version>/<config>.config 

After that we need to match the reco taus to the L1 taus. To do so edit the MakeTreeForCalibration.C file (inside the MatchAndCompress folder) with the correct in and out files, then just run:

root -l
.L MakeTreeForCalibration.C+
MakeTreeForCalibration()

After the matching the compression needs to be performed. To do so edit the produceTreeWithCompressedVars.py file (inside the MatchAndCompress folder) with the correct in and out files, then just run:

python produceTreeWithCompressedVars.py

Calibration

Enter Calibrate/RegressionTraining and run make clean ; make.

To do the calibration first create/modify the needed .config file inside the Calibrate/RegressionTraining/run directory, then just run:

./regression.exe run_<year>/<config>.config

Now that the regression has been trained, adapt to your needs the makeTH4_LUT.py file and run:

python makeTH4_LUT.py

Now that the TH4 LUTs have been created, we can apply the calibration to teh L1 objects. Adapt to your needs ApplyCalibration.C and ApplyCalibrationZeroBias.C

root -l
.L ApplyCalibration.C+
ApplyCalibration() # insert needed arguments

and

root -l
.L ApplyCalibrationZeroBias.C+
ApplyCalibrationZeroBias() # insert needed arguments

After the TH4 histos LUTs have been created we can make the LUTs that then go online; adapt to your needs the MakeTauCalibLUT.C file and run:

root -l
.L MakeTauCalibLUT.C+
MakeTauCalibLUT() # insert needed arguments

Isolation

Now that the calibration has been done the isolation needs to be compute and applied. To do so go to the Isolate folder.

To compute the isolation, adapt to your needs the Build_Isolation_WPs.C file and run:

root -l
.L Build_Isolation_WPs.C+
Build_Isolation_WPs() # insert needed arguments

This one has the possibility of building the WPs based on compressed or supercompressed variables. Due to statistics limits, it is always better to run in compressed mode.

Then the relaxation of the isolation needs to be performed. To do so, adapt to your needs Fill_RelaxedIsolation.C and run:

root -l
.L Fill_RelaxedIsolation.C+
Fill_RelaxedIsolation_TH3() # insert needed arguments

or adapt to your needs Fill_RelaxedIsolation_gridsearch_nTTextrap.C and run:

root -l
.L Fill_RelaxedIsolation_gridsearch_nTTextrap.C+
Fill_RelaxedIsolation_TH3() # insert needed arguments

to start a a grid search over the possible relaxation schemes that give rise to different turnON shapes. In here, the IsoEt cuts are constructed either by reading the bin's contents or making the IsoEt vs. nTT fit, the option regulating this is byBin.

Rates

To produce rates go to the MakeRates folder.

If you are running the 'old' version of the code, adapt to your needs Rate_ZeroBias_unpacked.C and Rate_ZeroBias.C and run:

root -l
.L Rate_ZeroBias_unpacked.C+
Rate() # insert needed arguments

and

root -l
.L Rate_ZeroBias.C+
Rate() # insert needed arguments

Else, if you are running the gridsearch, adapt to your needs Rate_ZeroBias_unpacked.C and Rate_ZeroBias_gridSearch.C and run:

root -l
.L Rate_ZeroBias_unpacked.C+
Rate() # insert needed arguments

and

root -l
.L Rate_ZeroBias_newnTT_gridSearch.C+
Rate() # insert needed arguments

Thresholds

Having computed the rates, the next step is to compute either the thresholds at fixed rates or the rates at fixed thresholds, by going to the CompareRates folder.

If you are running the 'old' version of the code, adapt to your needs CompareRates_ZeroBias_withUnpacked.C, and run:

root -l 
.L CompareRates_ZeroBias_withUnpacked.C
compare() # insert needed arguments

Else, if you are running the gridsearch, adapt to your needs CompareRates_ZeroBias_gridSearch_withUnpacked.C, and run:

root -l 
.L CompareRates_ZeroBias_gridSearch_withUnpacked.C
compare() # insert needed arguments

TurnONs

Now that also the isolation has been created we can test everything on the turn-on curves by going to the MakeTurnOns folder. Here the turnons can be made aother at fixed threshold or at fixed rates by applying the threshold computed at the previous step.

If you are running the 'old' version of the code, adapt to to your needs ApplyIsolationForTurnOns.C and run:

root -l
.L ApplyIsolationForTurnOns.C+
ApplyIsolationForTurnOns() # insert needed arguments

Else, if you are running the gridsearch, adapt to to your needs ApplyIsolationForTurnOns_gridSearch.C and run:

root -l
.L ApplyIsolationForTurnOns_gridSearch.C+
ApplyIsolationForTurnOns() # insert needed arguments

Gridsearch best options evaluation

If in the previous steps we have been using, the gridsearch approach to theoptimisation, we can now compare the different ptions to decide which one is the best for our needs.

To do so go into the CompareGridSearchTrunons, adapt to your needs BestFMturnOns_gridSearch.C, and run:

root -l 
.L BestFMturnOns_gridSearch.C
compare() # insert needed arguments

This one can compare the turnons at fixed threshold or at fixed rate. In both cases quality requirements are made on the turnon, and all information is saved in .txt files containing the optimisation figures of merit and the rates.

Validate performance on data

All of the abve has been done on MC and needs to be validated on data. Threforre, after having re-emulated the data with the new options we need to produce the turnons. We can do this in the PlotTurnOns folder.

We need adapting to your needs one of the three following codes:

• MakeEfficiencies_Data_reEmulated.C : make performance on re-emulated data with the re-optimised taus
• MakeEfficiencies_Data_unpacked.C : make performance on unpacked data from Run3 only
• MakeEfficiencies_Data_unpacked_withRun2.C : make performance on unpacked data from Run3 and Run2 at the same time and running:

root -l
.L MakeEfficiencies_Data_<tag>.C
compare() # insert needed arguments