Skip to content

Latest commit

 

History

History
140 lines (101 loc) · 6.55 KB

README.md

File metadata and controls

140 lines (101 loc) · 6.55 KB
Raw

Preprocessing module

Converts from root ntuples to PyTables files for DNN training.

  • variable transformation
  • weight calculation
  • ...

Setup

Option 1: Setup w/ LCG software stack (preferred)

For centos7 (e.g., LXPLUS):

source /cvmfs/sft.cern.ch/lcg/views/LCG_96bpython3/x86_64-centos7-gcc9-opt/setup.sh

Option 2: Setup w/ Miniconda (if LCG is not available)

Install miniconda if you don't have it:

wget https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh
bash Miniconda2-latest-Linux-x86_64.sh
# Follow the insturctions to finish the installation

Verify the installation is successful by running conda info.

If you cannot run conda command, check if the you added the conda path to your PATH variable in your bashrc/zshrc file, e.g.,

export PATH="$HOME/miniconda2/bin:$PATH"

Assume miniconda is installed at $HOME/miniconda2 on LXPLUS.

# create a new conda environment
conda create -n prep python=2.7

# set up ROOT
mkdir -p $HOME/miniconda2/envs/prep/etc/conda/
cd $HOME/miniconda2/envs/prep/etc/conda/
mkdir activate.d  deactivate.d
cd activate.d
# create the env_vars.sh file to get ROOT environment
cat << EOF > env_vars.sh
#!/bin/sh
# $HOME/miniconda2/envs/prep/etc/conda/activate.d/env_vars.sh
echo "Source root environment..."
# ROOT
source /cvmfs/sft.cern.ch/lcg/external/gcc/4.9.1/x86_64-slc6/setup.sh
cd /cvmfs/sft.cern.ch/lcg/releases/ROOT/6.07.06-7096a/x86_64-slc6-gcc49-opt/
source bin/thisroot.sh
cd -
EOF

# activate the environment
# NOTE: should not run any cmsenv beforehand
source activate prep

# verify ROOT is activated
which root
(/cvmfs/sft.cern.ch/lcg/releases/ROOT/6.07.06-7096a/x86_64-slc6-gcc49-opt/bin/root)

# install the necessary python packages
pip install numpy numexpr pandas scikit-learn scipy tables
pip install root-numpy

Instructions

Convert training/validation files

source activate prep

First activate the prep conda environment (if not yet done):

python runPreprocessing.py -n 50000 /eos/cms/store/path/to/input /eos/cms/store/path/to/output --data-format ak8_list --jobdir jobs -t condor

This will generate the condor submission script to convert the ntuples located at /eos/cms/store/path/to/input and output them to /eos/cms/store/path/to/output. Note that the input dir /eos/cms/store/path/to/input will be recursively searched and all files will be included, so make sure you put testing samples outside this directory!

  • It will first compute the metadata (i.e., the variable transformation, pT flattening weights, etc.) from the input root files. Note that this can take a long time so running with tmux or screen is recommended. The metadata will be saved in the output directory as metadata.json and can be re-used in the future (e.g., for converting the testing samples).
  • The data format (e.g., what branches to include, reweighting method, whether to make jet images, etc.,) is specified by the --data-format option. It should point to the python config file under preprocessing/data_formats (but without the .py suffix).
  • -t option sets the job type (condor or interactive). For condor submission, the submit script is generated but you need to run the condor_submit [your-submission-script] command to actually submit the jobs.
  • --jobdir opetion sets the directory for job-related files (submission script, logs, etc.). Set different job dirs if you are runnning multiple jobs at the same time.
  • -n option sets the number of events for each output file. The default value, 50000, is good for the nominal data format (pfcand list, or image).
  • If some of the jobs failed in condor, you can generate a resubmission script with only the failed jobs by invoking the --resubmit option:
python runPreprocessing.py -n 50000 /eos/cms/store/path/to/input /eos/cms/store/path/to/output --data-format ak8_list --jobdir jobs -t condor --resubmit

Convert testing files (JMAR samples)

For evaluating the performance and make the ROC curves, the samples specified by JMAR (https://twiki.cern.ch/twiki/bin/view/CMS/JetMETHeavyResPaper) are used. Since a specific sample is used for each signal category (Top/W/Z/H), each sample needs to be converted separately.

Note: the same metadata json file as used in converting the training file must be used for converting the testing files!

Take the example of converting the top sample:

# create a separate directory for the top sample
mkdir -p /eos/cms/store/path/to/output/JMAR/Top
# copy the metadata file used for converting the training dataset
cp /eos/cms/store/path/to/output/metadata.json /eos/cms/store/path/to/output/JMAR/Top
# create the jobs
python runPreprocessing.py -n 50000 /eos/cms/store/path/to/JMAR_sample/JMAR/ZprimeToTT_M-3000_W-30_TuneCUETP8M1_13TeV-madgraphMLM-pythia8 /eos/cms/store/path/to/output/JMAR/Top --data-format ak8_list --jobdir jobs_Top --remake-filelist
  • Specify the path for the Top sample as the input path, and output the files to the Top directory.
  • --remake-filelist option is needed to update the input file list using the specified input directory (otherwise the input files in the metadata file will be used).
  • --jobdir opetion sets the directory for job-related files (submission script, logs, etc.). Set different job dirs if you are runnning multiple jobs at the same time.

Reference preprocessing command

For nominal tagger training (94X, 20190326)

python runPreprocessing.py -n 200000 /eos/cms/store/cmst3/group/deepjet/ak8/ntuples/94X/20190326_ak8_links /eos/cms/store/cmst3/group/deepjet/ak8/hqu/20190326_ak8/ak8puppi_parts --data-format "ak8_list" --jobdir ak8puppi_parts_20190326 &> ak8puppi_parts_list_20190326.log &

For decorrelated tagger training (94X, 20190326)

# start with the same json to reuse the preprocessing parameters (median, lower/upper ranges)
# this would allow us to use the same testing files for both nominal tagger and the mass-decorrelated tagger
mkdir /eos/cms/store/cmst3/group/deepjet/ak8/hqu/20190326_ak8/ak8puppi_parts_ptmasswgt
cp /eos/cms/store/cmst3/group/deepjet/ak8/hqu/20190326_ak8/ak8puppi_parts/metadata.json /eos/cms/store/cmst3/group/deepjet/ak8/hqu/20190326_ak8/ak8puppi_parts_ptmasswgt

python runPreprocessing.py -n 200000 /eos/cms/store/cmst3/group/deepjet/ak8/ntuples/94X/20190326_ak8_links /eos/cms/store/cmst3/group/deepjet/ak8/hqu/20190326_ak8/ak8puppi_parts_ptmasswgt --data-format "ak8_list_ptmasswgt" --jobdir ak8puppi_parts_ptmasswgt_20190326 --remake-filelist --remake-weights &> ak8puppi_parts_list_ptmasswgt_20190326.log &