Getting started

This guide trains an interaction network (IN) on the CERN Open Dataset (http://doi.org/10.7483/OPENDATA.CMS.JGJX.MS7Q) to identify H->bb jets versus QCD background jets.

Installing

Checkout the prebuilt Docker containers.

  • If you use a GPU for training:

docker pull jmduarte/hbb_interaction_network:gpu

  • If you use a CPU for training:

docker pull jmduarte/hbb_interaction_network:cpu

Or rebuild the images from the Dockerfiles accordingly.

  • If you use a GPU for training:

docker build -f Dockerfile.gpu .

  • If you use a CPU for training:

docker build -f Dockerfile.cpu .

Run the image. Inside of the Docker container, clone the hbb_interaction_network git repository and install it.

git clone git@github.com:FAIR4HEP/hbb_interaction_network.git
cd hbb_interaction_network
pip install -e .

Convert dataset

To convert the full training dataset

python src/data/make_dataset.py --train

and the testing dataset:

python src/data/make_dataset.py --test

Training

To run the nominal training on CPU (or replace device with cuda to run on GPU):

python src/models/train_model.py --batch-size 1024 --epoch 100 --device cpu

Testing

To test the trained model:

python src/models/predict_model.py --batch-size 1024