In particle physics experiments, simulating calorimeter showers is computationally demanding, affecting the efficiency and accuracy of these experiments. Generative Machine Learning (ML) models have improved and sped up traditional physics simulation processes, but their use has largely been limited to fixed detector readout geometries. CaloPointFlow introduced a pioneering model capable of generating a calorimeter shower as a point cloud. This study advances the field with CaloPointFlow II, featuring several significant enhancements over its predecessor. These improvements include a novel dequantization technique, named CDF-Dequantization, and a new normalizing flow architecture, dubbed DeepSet-Flow. The updated model was tested using the fast Calorimeter Simulation Challenge (CaloChallenge) Dataset II and III, demonstrating its effectiveness and potential applications in particle physics simulations.
- CDF-Dequantization: A novel technique that improves the quality of generated point clouds by effectively handling the quantization of data, leading to more accurate simulations.
- DeepSet-Flow Architecture: A state-of-the-art normalizing flow architecture that efficiently processes set-structured data, enhancing the model's ability to generate complex calorimeter showers.
- Enhanced Efficiency: Significant improvements in computational efficiency, enabling faster simulations without sacrificing accuracy.
- Versatile Application: Tested with the CaloChallenge Dataset II and III, showcasing its applicability to a wide range of particle physics experiments.
- Clone the repository:
git clone https://github.com/simonschnake/CaloPointFlow.git
cd CaloPointFlow- Install the required Python packages, we use
python 3.10.5andpip:
python3.10 -m venv venv
source venv/bin/activate
pip install torch==2.2.2 torchvision==0.17.2 torchaudio==2.2.2 --index-url https://download.pytorch.org/whl/cu121
pip install torch-scatter -f https://data.pyg.org/whl/torch-2.2.2%2Bcu121.html
pip install click==8.1.7 zuko==1.1.0 tqdm==4.66.4 omegaconf==2.3.0 numba-0.59.1 pytorch-lightning==2.3.2 matplotlib==3.9.1 h5py==3.11.0 scipy==1.14.0 mplhep==0.3.50 tensorboardX==2.6.2.2
pip install .To train the model use
calopointflow train --help
Usage: calopointflow train [OPTIONS] [KWARGS]...
Options:
-m, --model TEXT Model to use. Options: I, dsf, dsf_cdeq, II
[required]
-d, --dataset INTEGER Dataset to use. Options: 2, 3 [required]
-ld, --log_dir TEXT Path to save the logs [required]
--help Show this message and exit.To generate new data use
calopointflow generate --help
Usage: calopointflow generate [OPTIONS] [KWARGS]...
Options:
-m, --model TEXT Model to use. Options: I, dsf, dsf_cdeq, II
[required]
-d, --dataset INTEGER Dataset to use. Options: 2, 3 [required]
--ckpt_path TEXT Path to the model checkpoint file [required]
--save_path TEXT Path to save the generated data [required]
--help Show this message and exit.To plot the histograms
calopointflow plot --help
Usage: calopointflow plot [OPTIONS]
Options:
-d, --dataset INTEGER Dataset to use. Options: 2, 3 [required]
-p, --plot TEXT What to plot.
Options:
all: Plot all available plots
marginals: Plot the marginal distributions
of the data in the z, alpha, and r
dimensions
layer_energies: Plot the energy
distributions of the data in individual
layer areas
corrcoeff: Plot the correlation
coefficients between the data in the z,
alpha, and r dimensions
cov_eigenvalues: Plot the histograms of
eigenvalues of the covariance matrices of
the individual showers
means: Plot the shower means in the z,
alpha, and r dimensions
cell_energies: Plot the energy
distributions of the data in individual
cells
num_hits: Plot the histogram of number of
hits
--save_path TEXT Path to save the plots
-g4, --geant4_data TEXT Path to the Geant4 data [required]
-cpf, --calopointflow_data TEXT
Path to the CaloPointFlow data [required]
--help Show this message and exit.This project is licensed under the MIT License - see the LICENSE file for details.
- Fast Calorimeter Simulation Challenge (CaloChallenge) for providing the datasets.