matmllab/sen_model

SEN Model

This is a repository for SEN model. Detailed description is prepared in Cap_Models file.

Description

This is a deep learning model for structure symmetry description, material property prediction, and material design, which is based on the symmetry-enhanced capsule transformer.

Development Environment

Package dependencies

• Please make sure about your running enviornment and package version.
• Linux (ubantu)>=18.04
• pymatgen==2022.4.19
• matplotlib>=3.0.3
• pandas>=1.0.5
• numpy>=1.16.2
• scikit_learn>=0.20.4
• keras==2.8.0
• sonnet==1.35
• tensorflow==2.8.0
• tensorflow_probability==0.8.0
• tensorflow_datasets==1.3.0

Package Installation

We can install all neccessary packages according to 'pip' or 'conda' with short time. All data can be downloaded from Material Project (https://materialsproject.org/). All data involved in this work will be available upon request. If you need full data, you can download from Material Project website via data/data.py. Or please contact us and provide the transmission address or email that can receive large files, and we will send you complete data.

About the Code

• The training and testing process are defined in main.py.
• The feature extraction block is defined in cap_block.py.
• The symmetry perception block is defined in models/che_env_block.py.
• Related data is in data file, and we also prepared the code data/data.py to get the dataset from Material Project.
• The capsule transformer is defined in cap_block.py.
• Related models are defined in models file.
• The visualization process and result plots in the manuscript are in plot file.
• The running time of one epoch with one batch is less than 1s under NVIDIA GV-100 GPU.

License

This project is covered under the Apache 2.0 License.

Thanks for your time and attention.

References

• Atz, K., Grisoni, F. & Schneider, G. Geometric deep learning on molecular representations. Nat. Mach. Intell. 3, 1023–1032 (2021).
• Kosiorek, A. R., Sabour, S., Teh, Y. W. & Hinton, G. E. Stacked capsule autoencoders. Preprint at http://arXiv.org/abs/1906.06818 (2019).
• Xie, T. & Grossman, J. C. Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties. Phys. Rev. Lett. 120, 145301 (2018).
• Chen, C., Ye, W., Zuo, Y., Zheng, C. & Ong, S. P. Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals. Chem. Mater. 31, 3564–3572 (2019).
• Liang, C., Jiang, H., Lin, S., Li, H., & Wang, B. Intelligent generation of evolutionary series in a time‐variant physical system via series pattern recognition. Adv. Intell. Sys.(2020)