alexdemoor / deepjet

DeepJet: Repository for training and evaluation of deep neural networks for Jet identification

This package depends on DeepJetCore 3.X (https://github.com/DL4Jets/DeepJetCore). (but it should still work for DJC2)

Usage

After logging in and setting up the DeepJetCore singularity environment, please source the DeepJet environment (please cd to the directory first!).

cd <your working dir>/DeepJet
source env.sh

The preparation for the training consists of the following steps

• define the data structure for the training. The DeepJet datastructure is found in the modules directory as the class TrainData_DF. DeepCSV is found as the class TrainData_DeepCSV

• convert the root file to the data strucure for training using DeepJetCore tools. The class argument should be TrainData_DF for DeepJet and TrainData_DeepCSV for DeepCSV:

    convertFromSource.py -i /path/to/the/root/ntuple/list_of_root_files.txt -o /output/path/that/needs/some/disk/space -c TrainData_DF
  

  This step can take a while.

• prepare the training file and the model. Please refer to DeepJet/Train/train_DeepFlavour.py for training of DeepJet and DeepJet/Train/train_DeepCSV.py for DeepCSV.

Training

Since the training can take a while, it is advised to open a screen session, such that it does not die at logout.

ssh lxplus.cern.ch
<note the machine you are on, e.g. lxplus058>
screen
ssh lxplus7

Then source the environment, and proceed with the training. Detach the screen session with ctr+a d. You can go back to the session by logging in to the machine the session is running on (e.g. lxplus58):

ssh lxplus.cern.ch
ssh lxplus058
screen -r

Please close the session when the training is finished

the training is launched in the following way:

python3 train_DeepFlavour.py /path/to/the/output/of/convert/dataCollection.djcdc <output dir of your choice>

If you use the PyTorch version for training ParticleTransformer AK4, please be aware this is a dev and unstable version. You will need to modify files depending on your will. The PyTorch training is launched in the following way:

python3 /path/to/DeepJet/ParT/Part_train.py /path/to/the/output/of/convert/dataCollection.djcdc <output dir of your choice>

Evaluation

After the training has finished, the performance can be evaluated.

predict.py <output dir of training>/KERAS_model.h5  <output dir of training>/trainsamples.dc <dir with test sample stored as rootfiles>/filelist.txt <output directory>

This creates output trees with the prediction scores as well as truth information and some kinematic variables.

Similarly to the training, the PyTorch prediction The PyTorch prediction is launched in the following way:

python3 /path/to/DeepJet/ParT/Part_predict.py ParticleTransformer <output dir of training>/checkpoint.pth  <output dir of training>/trainsamples.djcdc <dir with test sample stored as rootfiles>/filelist.txt <output directory>

Condor batch GPUs

For training a NN using condor's GPUs we mostly use interactive sessions. You will need 3 files available in DeepJet-pytorch fork (sing.sh + sing.sub for condor and run_deepjetcore3.sh for our Singularity container).

The commands for starting a GPU session are the next ones :

condor_submit -interactive /path/to/sing.sub
source /path/to/run_deepjetcore3.sh
export KRB5CCNAME=/tmp/cred1111
kinit

You are now ready to train/predict any NN you want !