Generalizable collider physics generator-level event reconstruction with a graph DNN.
1. Clone this repository
git clone git@github.com:sam-may/EverGraphDNN.git
cd EverGraphDNN
2. Install dependencies
The necessary dependencies (listed in environment.yml) can be installed manually, but the suggested way is to create a [conda environment](https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/mana
ge-environments.html) by running:
conda env create -f environment.yml
3. Install evergraph
Suggested way to install is:
pip install -e .
An output file from HiggsDNA after running the selection in higgs_dna/evergraph_tagger.py is available in this directory: /home/users/smay/HiggsDNA/scripts/evergraph_25Jan2022/.
To prep inputs for DNN training with a hadronic selection (>=4 jets, 0 leptons + taus) and selecting ttH and ttHH->ggbb, run the command:
python scripts/prep.py --input_dir "/home/users/smay/HiggsDNA/scripts/evergraph_25Jan2022/" --output_dir "hadronic_1Mar2022" --log-level "DEBUG" --selection "Hadronic" --objects "photons,jets,met"
and to train a graph CNN which performs convolutional operations on each pair of input objects from the 2 photons, 8 leading jets, and MET (11 choose 2 = 55 pairs per event), run the following command:
python train.py --input_dir "hadronic_1Mar2022/" --output_dir "hadronic_1Mar2022_tthh_vs_tth/" --log-level "DEBUG"
this graph CNN will be trained with the following targets:
- whether the event has an H->bb pair or not (
"target_has_HbbHiggs")
and optionally you can add other targets through evergraph/algorithms/dnn_helper.py, e.g.:
- the pt and eta of the H->bb Higgs
- the pt and eta of the H->gg Higgs (as a sanity check we can compare this with the pt and eta we get from adding the four vectors of the photons)
TODO items:
- Make training targets and network details configurable through
json - Develop
evergraph/evaluationtools for assessing performance of graph CNNs