MiniAOD_photons_to_ML

This code extracts photon candidates from CMS MiniAOD files and is meant to build datasets to train deep-learning models to separate real photons from fakes.

The files CRAB_config.py and slim_MiniAODs.py are needed to slim the MiniAOD files and store locally (already done).

For the further analysis of the data, MiniAOD_analyser.py can be used.
How to run the code:

1) Setup CMSSW environment

Further information: https://twiki.cern.ch/twiki/bin/view/CMSPublic/WorkBookSetComputerNode

Build a new directory where you want to do your analysis and run the following commands:

source /cvmfs/cms.cern.ch/cmsset_default.sh

cmsrel CMSSW_12_6_0

cd CMSSW_12_6_0/src

cmsenv

2) Clone repository and run the code

git clone https://github.com/FloMau/MiniAOD_photons_to_ML.git (or better: fork it and clone your fork!)

cd MiniAOD_photons_for_ML

Now you are ready to go! You can run the script using

python3 MiniAOD_analyser.py

After first time with installing, you can skip some of the steps above, just do:

source /cvmfs/cms.cern.ch/cmsset_default.sh

cd CMSSW_12_6_0/src

cmsenv

cd MiniAOD_photons_for_ML

python3 MiniAOD_analyser.py

ToDo

This code is still under development and contains only a first skeleton at this point.

• Implement CMS-standard pre-selection cuts on photon candidate shower shapes which are applied before the MVA
• Make the code save the ECAL RecHits, separately for real and fake photons in an appropriate format (e.g. using numpy) for the ML studies that we want to do
• Validation steps: plotting of shower shapes distributions and calorimeter images seperately for real photons and fakes
• Add tracks to our calorimeter images / graphs

Further information about how to read MiniAOD files: https://twiki.cern.ch/twiki/bin/view/CMSPublic/WorkBookMiniAOD2017#4_7_MiniAOD_Analysis_Documentati

For tracks and isolation, this will be helpful: https://github.com/cms-sw/cmssw/blob/master/RecoEgamma/PhotonIdentification/plugins/PhotonIDValueMapProducer.cc#L254