Purpose of this model is to classify 4 particle types (pions, electrons, protons, kaons) using sparse CNNs. Minkowski Engine is used for sparse convolutions.
a) Minkowski Engine
b) scikit-learn (optional).
You will require eic-shell (https://eic.phy.anl.gov/tutorials/eic_tutorial/getting-started/quickstart/).
Enter eic-shell:
./eic-shell
Enter nightly:
source /opt/detector/setup.sh
Generate data, as an example:
npsim --compact epic.xml --enableGun --gun.particle "e-" --gun.energy "5*GeV" --gun.thetaMin "3*deg" --gun.thetaMax "50*deg"
--gun.phiMin "50*deg" --gun.phiMax "85*deg" --gun.distribution "cos(theta)" --numberOfEvents 100000
--outputFile e-_5GeV_20deg_22deg_1e5.edm4hep.root
clean.py is used to filter data in an appropriate format for Minkowski Engine. You will need to write your own cleaning code to use the dense.py code.
sparse.py is used, support for Bayesian hyperparameter optimization is provided in the code using scikit-learn.