HEPHeroML - Machine Learning tool for the DESY-CBPF-UERJ collaboration

General information

• This code is meant to be used in association with the HEPHero framework.

• The training setup is made in the python scripts inside the directory setups.

• The input data consists of h5 files created by the tool grouper.py of the HEPHero framework.

• The training results and files are stored in the smae directory of the h5 files.

Starting

Inside your private area (NOT in the eos or dust area and NOT inside a CMSSW release), download the code.

git clone https://github.com/DESY-CBPF-UERJ/HEPHeroML.git

Source the hepenv environment before work with the HEPHeroML:

hepenv

Enter in the HEPHeroML directory:

cd HEPHeroML

Generating the trainer

After setup the model and training in one of the python scripts inside setups, generate the trainer using the generate_trainer.py script. Example: Generate the trainer for the analysis OPENDATA with the tag Class (defined inside setups/OPENDATA.py):

python generate_trainer.py -a OPENDATA

It will create the trainer script train_OPENDATA_Class.py.

Running the trainer

Know how many jobs the code is setted to train (information needed to submit jobs):

python train_OPENDATA_Class.py -j -1

List the jobs the code is setted to train:

python train_OPENDATA_Class.py -j -2

Train the model in the position n of the list:

python train_OPENDATA_Class.py -j n

Ex.:

python train_OPENDATA_Class.py -j 2

Submit condor jobs

1. Make submit_jobs.sh an executable:

chmod +x submit_jobs.sh

2. See all flavours available for the jobs:

./submit_jobs.sh help

3. Submit all the N jobs the code is setted to train:

./submit_jobs.sh flavour N trainer

Ex.:

./submit_jobs.sh workday 32 train_OPENDATA_Class.py

Evaluate the results

After the jobs have finished, evaluate the training results:

python evaluate.py -s selection_name -p period -a analysis -t tag -l library

Ex.:

python evaluate.py -s MLOD -p 12 -a OPENDATA -t Class -l torch

period = 12, APV_16, 16, 17, or 18