This code is based in the Run2 Analysis https://github.com/irenedutta23/HmmAna

Setup

cmsrel CMSSW_15_0_19
cd CMSSW_15_0_19/src
cmsenv
git clone git@github.com:LPC-HH/HToMuMu.git
scram b clean
scram b -j 8
cd HToMuMu
make -j4

For jobs and dataset creation

You need to have a CMS GRID certificate established

voms-proxy-init -voms cms

Run the analyzer

The format for running the analyzer is

./bin/HmmAnalyzer runList.txt(or .list) out.root mc/data is_data year_num if_include_Z_region if_include_H_region

For simulations

./bin/HmmAnalyzer runList.txt out.root mc F 2024 F T

For data

./bin/HmmAnalyzer runList.txt out.root data T 2024 F T

To run over all the data sets look at condor/README.md to run it in condor jobs

Run CreateTuples

This will create a root file with tuples with all the usefull variables and the correspodeting weights;

For data/simulation

./bin/CreateTuple analyzer_output.root output_directory era dataset T(data)/F(simulation) T(signal)/F(Bkg-data)

To run over all the data sets look at condor/README.md to run it in condor jobs.
If you want to run it over a small dataset look at scripts/README.md

Using Condor jobs

Analyzer and tuplizer should be run over all datasets using condor jobs. To do this, look at the README in condor/

Hadd Tuplizer output

After you have the tuples, you need them locally to plot from and run the BDT on.

cd ./root_io/tuples/
bash hadd_tuples_EOS.sh
bash hadd_tuples.sh

The second hadd creates combined era tuple files

Skim tuples

BDT does not like big files, so we need to generate a tuples that contain just the events from that production mode and the training variables. To this use src/SkimTuples_VBF.cc src/SkimTuplesggH.cc and the scripts ./scripts/run_skim_ggH.py ./scripts/run_skim_VBF.py You can change the datasets by changing the lists eras, background_datasets, and signal_datasets in the script.

Run

python3 ./scripts/run_skim_ggH.py
python3 ./scripts/run_skim_VBF.py

Once we have the skim tuples, you must hadd to have one background tuple and one signal sample, you can just do

cd ./root_io/skim/(ggH/VBF)
bash hadd_skim.sh

You can also run the skimming, hadd, and removal of previous tuples at once with:

bash scripts/run_skim.sh (bdt/kin/both)

This logs the scripts for the skimming and hadd in scripts/logs/

BDT Training

Look at python/xgboost/README.md

Plotting

All plotting codes are located at plot_macros/

How to Run the Complete Analysis

This section describes the complete workflow, from producing the analyzer tuples to calculating the final systematic uncertainties.

1. Set the analysis version

Go to the condor/ directory:

cd ./condor/

Edit:

./condor/utils/version.py

and set the version name you want to use to tag your tuples.

---

2. Produce the analyzer tuples

From the condor/ directory, run:

python3 create_analyzer_jobs_LPC.py
bash condor_job_sender.sh

This creates and submits the analyzer jobs. The resulting tuples will be stored in the EOS space defined by the configuration.

Optional: produce muon (p_T) systematic tuples

If you also want to produce the muon (p_T) systematic variations, run:

python3 create_analyzer_jobs_LPC_muonsys.py
bash condor_ctuple_job_sender_syst.sh

---

3. Check the analyzer jobs

Once the jobs have finished, check for failed or missing jobs with:

bash analyzer_job_checker_rerun.sh version

Replace "version" with the analysis version you are using. For example:

bash analyzer_job_checker_rerun.sh 2.2

The script will check the submitted jobs and rerun the ones that failed.

---

4. Compute the sum of generated weights

To produce the histograms containing the sum of generated weights, go to the scripts/ directory:

cd ../scripts/

Run:

python3 ComputeWeights.py
python3 ComputeWeights_muon_sys.py

ComputeWeights.py computes the nominal generated-weight information, while ComputeWeights_muon_sys.py does the same for the muon systematic samples.

---

5. Produce the tuples

Return to the condor/ directory:

cd ../condor/

Create and submit the nominal ctuple jobs:

python3 create_ctuples_jobs_LPC.py
bash condor_ctuple_job_sender.sh

Optional: produce muon and jet (p_T) systematic ctuples

To produce the ctuples for the muon and jet (p_T) systematic variations, run:

python3 create_ctuples_jobs_LPC_syst.py
bash condor_ctuple_job_sender_syst.sh

---

6. Check the tuple jobs

After the jobs have finished, check for failed jobs with:

bash ctuple_job_checker_rerun.sh version

For example:

bash ctuple_job_checker_rerun.sh 2.2.2

---

7. Merge (hadd) the nominal tuples

Go to the tuple directory:

cd ../root_io/tuples/

Run:

bash hadd_tuples_EOS.sh -eos
bash hadd_tuples

This merges the nominal tuples required for the analysis.

---

8. Merge the systematic tuples

To merge the systematic tuples using Condor jobs, return to the condor/ directory:

cd ../../condor/

Then run:

bash submit_hadd_systematics.sh

---

9. Add the BDT scores to the systematic tuples

Go to the XGBoost directory:

cd ../python/xgboost/

Submit the jobs that add the BDT scores to the systematic tuples:

bash submit_bdt_systematics.sh

---

10. Calculate the systematic uncertainties

Finally, return to the initial project directory and run:

bash ./scripts/calculate_all_systematics.sh

c This calculates the systematic uncertainties from the nominal and systematic tuples.

---

Workflow summary

The complete workflow is:

Set version
    ↓
Produce analyzer tuples
    ↓
Check analyzer jobs
    ↓
Compute generated-weight histograms
    ↓
Produce ctuples
    ↓
Check ctuple jobs
    ↓
Hadd nominal tuples
    ↓
Hadd systematic tuples
    ↓
Add BDT scores to systematic tuples
    ↓
Calculate systematic uncertainties

The muon and jet systematic steps are optional if only the nominal analysis is needed.