This package provides following utilities:

• Predict total energies of crystal strtuctures using a pre-trained model
• Train a model with a customized dataset

Follwoing python packages needs to be installed:

• TensorFlow 2
• TensorFlow Addons
• pymatgen
• scikit-learn
• nfp

• Predict total energy using a pre-trained model
• Train a model using a customized dataset
• Pre-trained models

To reporduce the models in our paper, following data can be downloaded:

• Total Energy

  • Total energies of ICSD structures from NRELMatDB databse (materials.nrel.gov)
  • Total energies of hypothetical structures

• Crystal Structures

  • ICSD crystal structure files cannot be distributed
  • Relaxed hypothetical crystal structures
  • Unrelaxed hypothetical structures

• Shubham Pandey ([email protected])
• Peter St. John ([email protected])
• Prashun Gorai ([email protected])

"Predicting energy and stability of known and hypothetical crystals using graph neural network"

S. Pandey, J. Qu, V. Stevanovic, P. St. John, and P. Gorai, Patterns (2021). DOI: 10.1016/j.patter.2021.100361

This package is released under the MIT License.