DOSMatGen

KO | EN

gitlite — search

#python #hacktoberfest #react #ai #llm #android #rust #linux #docker #nodejs #windows #vue #ios #mcp

(★ 14)

Electronic Structure Guided Inverse Design Using Generative Models

DOSMatGen 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# DOSMatGen
[![arXiv](https://img.shields.io/badge/arXiv-1234.56789-b31b1b.svg)](https://arxiv.org/abs/2504.06249)
Electronic Structure Guided Inverse Design Using Generative Models

## Environment
```
# create a conda environment
conda create -n dosmatgen python=3.12
conda activate dosmatgen

# pytorch
conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia

# pytorch geometric
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.4.0+cu124.html

# install additional dependencies
pip install ase pymatgen lightning p-tqdm omegaconf einops wandb

# ipynb support
pip install ipykernel ipywidgets
```

## Dataset
The datamodule expects crystal data in JSON format, structured as a list of dictionaries with the following keys:
```
[
    ...
    {
        'structure_id'      : str,
        'positions'         : list of (N, 3),
        'cell'              : list of (3, 3),
        'atomic_numbers'    : list of (N),
        'y'                 : list of (N, M)
    }
    ...
]
```
where `y` is the node-level features.

See below on how to download the MP-DOS dataset.

## Sample Run
To train a diffusion model with CFG (classifier-free guidance) on the MP-DOS dataset, first take a look at the config file `configs/dos_cfg.yml`, then run
```
python run_diffusion_CFG.py
```
The training progress is tracked using `wandb`. If you're running it for the first time, you may need to sign in.

## Generation
Since we are using classifier-free guidance, we are still able to do unconditional generation, that is, generating structures without input DOS information. To do so, try
```
python generate_CFG_unconditional.py --root_path=<PATH/TO/CKPT>
```

For conditional generation, we need to feed the model the DOS information we want to condition the generation on. As an example, we can try to reconstruct the MP-DOS test split:

```
python generate_CFG_conditional.py --root_path=<PATH/TO/CKPT>
```

Generated structures will be saved to a timestamped folder in `structures/`.

## Visualization
Please refer to `plot_dos.ipynb` on a possible approach to visualize the generated structures' DOS against the ground truth.

## Download Model Weights and Dataset
Two trained model weights, (1) diffusion model with CFG and (2) DOS forward model, can be download at the following link:
https://u.pcloud.link/publink/show?code=XZ0kE45ZOuEPUrHzQU5ciR2pXKCkDzNY3F9k

The MP-DOS dataset is available at the following link:
https://u.pcloud.link/publink/show?code=XZ4kE45ZhFgF7IWysbuKnuUExSsnfLC2n9wy

The default directory structure for this repo is:

```
DOSMatGen
├── configs
├── data
│   └── mp_dos
├── outputs
│   ├── dos_cfg
│   └── dos_forward_model
├── dosmatgen
│   ├── dataset
│   ├── diffusion
│   ├── models
│   └── utils
└── structures
    └── 08032025_160513
```

## Acknowledgements

This work builds upon prior research in generative modeling for crystal structure prediction. In particular, we acknowledge the contributions of DiffCSP, on which this work is informed on. We thank the authors for making their code and ideas publicly available.

## Citation
```
@article{jia2025electronic,
  title={Electronic Structure Guided Inverse Design Using Generative Models},
  author={Jia, Shuyi and Ganesh, Panchapakesan and Fung, Victor},
  journal={arXiv preprint arXiv:2504.06249},
  year={2025}
}
```

DOSMatGen

DOSMatGen 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

DOSMatGen

DOSMatGen 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README