CGNE (Crystal Growth Neural Emulator)

Official repository for the paper Efficient Probabilistic Modeling of Crystallization at Mesoscopic Scale: https://arxiv.org/abs/2405.16608

Requirements

CGNE

• Python 3.10
• PyTorch
• Lightning
• NumPy
• H5Py
• PyYAML (for configuration files)
• Wandb (for logging)
• einops (for logging)
• POT (for calculation of Wasserstein distance)

LCA

• CUDA compatible GPU
• Python 3.10
• NumPy
• Numba
• H5Py
• PyYAML
• Matplotlib

Usage

CGNE

To train the model, run the following command:

python -m CGNE.train --config config.yaml

where config.yaml is the configuration file. A configuration file with default values is provided in configs/default.yaml.

Any configuration parameter can be overridden by adding it to the configuration file. For example:

model:
  z_dim: 128
data:
  batch_size: 64

LCA

To run the LCA simulation, run the following command:

python -m LCA.main

This accepts command line arguments, as well as a configuration file.

Citation

If you use this code, please cite the following paper:

@misc{timmer2024efficient,
      title={Efficient Probabilistic Modeling of Crystallization at Mesoscopic Scale}, 
      author={Pol Timmer and Koen Minartz and Vlado Menkovski},
      year={2024},
      eprint={2405.16608},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}