D-CGCNN is a CGCNN (xie et al) based python code with direction-based crystal graph representation. D-CGCNN is intended to predict formation energies of relaxed structures using unrelaxed structures as inputs, where unrelaxed structures can be generated by a simple elemental substitution or generative models.
model.py is dropout-added graph convolutional neural network, modified version of the original CGCNN by Dr. Juwhan Noh, currently at KRICT.
For more details, check out this paper (https://pubs.acs.org/doi/pdf/10.1021/acs.chemmater.2c02498).
- Python3
- Numpy
- Pytorch
- Pymatgen
- Sklearn
- Clone thie repository:
git clone https://github.com/SeoinBack/D-CGCNN
- Run example to test
python main.py
python predict.py
- More details about inputs and arguments are same to original CGCNN, so you can found it in https://github.com/txie-93/cgcnn
If you use this code, please cite:
Dong Hyeon Mok, Jongseung Kim, Seoin Back. "Direction-Based Graph Representation to Accelerate Stable Catalyst Discovery" (https://pubs.acs.org/doi/pdf/10.1021/acs.chemmater.2c02498)