Skip to content

Latest commit

 

History

History
40 lines (25 loc) · 1.83 KB

README.md

File metadata and controls

40 lines (25 loc) · 1.83 KB
Raw

Caloscore official repository

In this repository, the implementation of the studies presented in the paper: Score-based Generative Models for Calorimeter Shower Simulation.

DOI

The score-based generative model is trained using a diffusion process that slowly perturbs the data. Generation of new samples is carried out by reversing the diffusion process using the learned score-function, or the gradient of the data density. For different time-steps, we show the distribution of deposited energies versus generated particle energies (top) and the energy deposition in a single layer of a calorimeter (bottom), generated with our proposed CaloScore model.

Tensorflow 2.6.0 was used to implement all models, based on the score-model implementation from Score-Based Generative Modeling through Stochastic Differential Equations

Data

Results are presented using the Fast Calorimeter Data Challenge dataset and are available for download on zenodo:

Run the training scripts with

cd scripts
python train.py  --config CONFIG --model MODEL

  • MODEL options are: subVPSDE/VESDE/VPSDE

  • CONFIG options are [config_dataset1.json/config_dataset2.json/config_dataset3.json]

Sampling from the learned score function

python plot_caloscore.py  --nevts N  --sample  --config CONFIG --model MODEL

Creating the plots shown in the paper

python plot_caloscore.py  --config CONFIG --model MODEL --nslices 1