anaTprimeHad

Searching for Tprime in Hadronic channel

Analysis workflow

Setup tested in cca.in2p3.fr

• pyroot, xrootd, slurm needed

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

cmsrel CMSSW_14_1_12
cd CMSSW_14_1_12/src
cmsenv
git clone https://github.com/sarakm0704/anaTprimeHad.git
cd anaTprimeHad/ana
make clean
make -j 30

to test

mkdir test
./processnanoaod.py filelist/list_Tprime600UL18.txt test/Tprime600 tprimeConfig_UL18_3T > test/tp600.out

for htop server (if sample exists locally)

./processnanoaod.py filelist/htoplist_TThad_test.txt test/TTToHadronic ttConfig_UL18_3T > test/tt.out

This workflow is currently using nanoaod via xrootd and running over slurm only. It could be modified to use HTCondor.

To run with complete datasets:

• modify tasks in sjob.sh
  • -L sps option is only allowed in cca.in2p3.fr to use sps storage
• individual tasks are defined in shell_slurm/ directory
  • but options for slurm job such as --chdir, --mail*, X509_USER_PROXY and storage locations should be modified according to the analyser themselves

After defining options correctly, run the job

./sjob.sh

NN workflow

Treating arrays for training inputs / prediction outputs

To convert training ntuples to hdf:

./tree2hdf.py -d -p

• -d: convert single root file to hdf
• -p: merge single hdf files per process

then to cut out training datasets for balancing training inputs into certain fractions and shuffle:

python3 shuffle_mergy.py

to convert hdf output with prediction scores into TTree once training is done:

./hdf2tree.py

Training and evaluation

Training has been performed DNN+CNN architecture with train_cnn.py

python3 train_cnn.py

To run over all the samples and save prediction scores (e.g. data) use evaluation

cnn_evaluation.py -e

input/output location needs to be specified when train/evaluate using code aboves

There is a tool to have evaluation, correlation matrices, feature importance and input distribution of features. Dedicated for further studies only.

python doFeatures.py -e -i -c -d (or -t)

• -e: do Evaluation of model
• -i: Extract Feature Importance
• -c: Extract Correlations
• -d: Draw Input variable Distributions
• -t: Test with small portion from given dataset (100 events)

keras (2.8.0), pyroot (6.30/02) used above

nanoAOD private production

ref: https://gitlab.cern.ch/cms-nanoAOD/nanoaod-doc/-/wikis/Instructions/Private%20production

Don't forget to source the environment from cvmfs if you are lyoserv user by doing:

cms_env

With 2018 datasets for example:

cmsrel CMSSW_10_6_27
cd CMSSW_10_6_27/src
cmsenv
git clone https://github.com/sarakm0704/anaTprimeHad.git
cd anaTprimeHad/prod
crab submit -c crab_mc.py