Recent developments in calorimeter physics are leading to a new paradigm in calorimeters' design. The aim is to reconstruct the whole spatial distribution of the shower instead of extracting information only about the energy deposition inside calorimeter cells. Many collaborations are designing highly-segmented calorimeters (for instance, CMS High-Granularity CAL and ALICE FoCAL) to reach this goal. In this project, our goal is to build a neural network to simulate electromagnetic shower energy deposition inside a toy segmented calorimeter. Taking inspiration from recent works in this field, like the CaloGAN network, we have built a conditional GAN with auxiliary conditions based on total energy deposition and particle IDs. The neural network was trained with a dataset created on purpose with the Geant4 simulation toolkit. Simulations involve electrons, positrons and photons with energies from 1 to 30 GeV that strike on a CsI calorimeter with 12 layers and 12x12 cells. The results are further analyzed with ROOT to evaluate GAN's performances. Due to time, dataset and hardware constraints, this project must be considered an exploratory work whose results can even be improved with more resources.
Installing from local source in Development Mode, i.e. in such a way that the project appears to be installed, but yet is still editable from the src tree:
$ git clone https://github.com/Dario-Maglio/EM-shower-simulator-with-NN.git
$ cd EM-shower-simulator-with-NN
$ python3 -m pip install -e .All of the package's informations are stored in setup.cfg and are passed as arguments to setuptools.setup() when it is executed through pip install.
If no arguments are passed, the simulation will use random features as input.
In python you can import the module simply writing
>>> import em_shower_simulator as emthen you can run the simulation module as
>>> em.simulate([10.0, 1], verbose=0)
simulating event with features [10.0, 1.0]
0This returns the error code and shows the resulting plot of the simulation.
You can also run the simulation from command line, writing:
$ simulate-EM-shower --features 10. 1This command allows the user to generate a shower simulation from command line. Furthermore, it allows to pass the shower's features as float arguments in the order: energy momentum angle Use --help for more information about the arguments:
$ simulate-EM-shower --help
usage: simulate-EM-shower [-h] [-v] [-f feature [feature ...]]
...It gives an error if a different input size or type is passed.
Output example of 8 particles. Can you recognize them?
This repository is licensed under the GNU General Public License v3.0 (GPL-3.0).
See the LICENSE file for more information.