HHH analysis framework

Framework to perform HHH analysis, run MVA tree inputs, histograms, apply calibrations and BDT evaluation and produce maps.

Getting started

Requires CMSSW with python3 enabled.

cmsrel CMSSW_12_5_2
cd CMSSW_12_5_2/src
cmsenv

Then, get the repository:

git clone [email protected]:mstamenk/hhh-analysis-framework.git
cd hhh-analysis-framework
source setup.sh

Then every time in a new shell:

cd hhh-analysis-framework
source setup.sh

Installing additional softwares

Need for calibrations:

git clone ssh://[email protected]:7999/cms-nanoAOD/jsonpog-integration.git

Necessary step: producing efficiency maps

Efficiency maps are necessary to create truth tagging, and to use b-tagging calibrations in the case of jets failing the b-tagging requirement. The maps need to be produced.

cd eff-maps
python compute_efficiency_maps.py # need to modify line 26 where path = '' is defined to use the nanoaod samples
python build_map.py

Alternative getting started

Instead of CMSSW you can also just pick up python and RDataframes in lxplus8 with

source /cvmfs/sft.cern.ch/lcg/views/setupViews.sh LCG_102 x86_64-centos7-gcc11-opt

Running the framework

Doing trees with baseline selection from nanoAOD (trigger and basic pt cuts)

@Marko, please add how you do the inclusive (with an explicit example of the options)

cd hhh-analysis-framework
python make_histograms_rdataframe_selection.py # see options

Scripts to b-tag and fat-jet categorization

With the script bellow we redefine a couple of variables and construct trees, histograms and data/MC plots to different categories

• several btag and fat jet categories selections are implemented. There is also an option to make the results with SR/CR separation

• read the script here for the category implemented_category options. Implemented more as needed (eg PN categorization in boosted categories)

python3 skimm_tree.py --category implemented_category --base_folder /eos/user/m/mstamenk/CxAOD31run/hhh-6b/v25/2017/baseline

• The histograms options are here.

  • If you change the histogram binning and want to redo the plots, there is no need to re-do tress, just pass --skip_do_trees as argument to the skimm_tree.py

• The plotting options are here.

  • If you change just plotting options (eg color, stack ordering, ...) and want to redo the plots, there is no need to re-do tress, just pass --skip_do_trees --skip_do_histograms as argument to the skimm_tree.py

Read the other options as necessary

One can also inspect the yields on the different categories with the bellow.

MVA training

• XGBoost based training scripts are here
• @Marko add instructions to TMVA

Instructions to run TMVA

First step is to fetch the data samples and save a train / test for signal and background.

This can be done in the prepare_datasets_boosted.py (also usable for resolved)

python prepare_datasets_boosted.py

In the script, one needs to specify the paths to the resolved or boosted samples and what selection to apply.

Once the training and testing sets a completed, one can use the script train_bdt_resolved.py to do the training.

python train_bdt_resolved.py --year run2 --nTrees 200 --maxDepth 3 --nCuts 50 --minNodeSize 5

Datacards maker

• See here