chrispap95 / deadCellRegression

deadCellRegression

Repository for ML study on dead Si cells rechit regression. For use with ROOT >= 6.20.00 (CMSSW_11_1_0)

Initial setup

Setup latest CMSSW release

cmsrel CMSSW_11_2_0_pre2
cd CMSSW_11_2_0_pre2/src
cmsenv

Get the repository

git clone https://github.com/chrispap95/deadCellRegression.git

cd to the area

cd deadCellRegression/regressionScripts

Train the algorithm

root -l TMVARegression.C\(\"someSample.root\",\"testRun\",10000\)

to use the result

root -l TMVARegressionApplication.C\(100,1\)

Explanation of main algorithms

To get an idea of what the inputs are:

root -l TMVARegression.C\(\"inputFile.root\",\"uniqueIDstring\",nSamples,nHiddenLayers,\"nodesPerLayer\"\)
root -l TMVARegressionApplication.C\(energy,deadFraction\)

The training script can be also submitted through condor using prepareCondor.sh,condor-exec.csh,condor.jdl. When the files are configured properly, submit using

condor_submit condor.jdl

Keras regression

Use deadCellsRegression*.py scripts for Keras based regression. You will need to add the input/output names in the scripts. Ultimate goal is to make these the default scripts for regression.

Fit and plot resolution

The script rechitSumLooper.C can be used for easy fitting of the discrete energy samples. Run it as follows:

root -l rechitSumLooper.C\(deadFraction,\"methodToUse\"\)

The available methods are none,MLregr,aver and LSaver. Finally, to combine multiple plots into one, you can use plotComparison.C. It uses the root files from the previous step as input.