Skip to content

Files and data needed to reproduce the machine learning results of the paper "Materials discovery using generative machine learning"

License

Notifications You must be signed in to change notification settings

newalexander/generative-materials-discovery

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 

Repository files navigation

materials-discovery

This repo contains files and data needed to reproduce the machine learning results of the paper "Materials discovery using generative machine learning". Further data is available at: https://doi.org/10.6084/m9.figshare.25783221.v4.

In particular, it contains a modified version of the 2023.01.10 release of the alignn package (https://github.com/usnistgov/alignn/tree/v2023.01.10), developed by NIST. This package is associated with the ALIGNN framework (Choudhary et al., 2021).

License

See the LICENSE.rst file in alignn/ for the license used by NIST for releasing alignn.

Scripts developed by JHU/APL have the following license:

Copyright 2024 Johns Hopkins University Applied Physics Laboratory

Licensed under the Apache License, Version 2.0

Instructions (PGCGM)

We used the pretrained PGCGM (Zhao et al., 2023) model provided in the current release (https://github.com/MilesZhao/PGCGM/tree/ab38ec6e4e6205253fb583ae0667b7a90e91cee5). Generating materials used the create_cif.py, pymatgen_valid.py, and merge_valid.py scripts in that repo.

Installation (ALIGNN)

Follow these steps to get the library working on a GPU-capable machine. Note that we've commented out the install_requires kwarg of setup.py because we're manually specifying dependencies prior to installing alignn.

cd ...  # navigate to this directory beforehand.
conda create --name alignn python=3.8
conda activate alignn
mamba install numpy scipy=1.6.1 scikit-learn=0.22.2 matplotlib=3.4.1 pandas=1.2.3 -c conda-forge
mamba install jarvis-tools=2021.07.19 -c conda-forge
pip install tqdm pydantic==1.10.7 cif2cell==2.0.0a3 flake8 pycodestyle pydocstyle pyparsing ase
pip install torch==1.12.0+cu116 --extra-index-url https://download.pytorch.org/whl/cu116
mamba install -c dglteam "dgl=0.9"
pip install pytorch-ignite==0.5.0.dev20221024
pip install -e alignn/

Contents

The two relevant scripts are:

  • scripts/train_stability.sh: Use the MP data to train an ALIGNN model to predict decomposition enthalpy.
  • scripts/Ed_tern_pgcgm_evaluation.sh: Use the trained ALGINN model to prediction decomposition enthalpy of generated structures.

References

  • Bartel, C. J., Trewartha, A., Wang, Q., Dunn, A., Jain, A., and Ceder, G. A critical examination of com- pound stability predictions from machine-learned for- mation energies. npj Computational Materials, 6(1): 97, Jul 2020. ISSN 2057-3960. doi: 10.1038/s41524-020-00362-y. URL https://doi.org/10.1038/s41524-020-00362-y
  • Choudhary, K. and DeCost, B. Atomistic line graph neural network for improved materials property pre- dictions. npj Computational Materials, 7(1):185, Nov 2021. ISSN 2057-3960. doi: 10.1038/s41524-021-00650-1. URL https://doi.org/10.1038/s41524-021-00650-1
  • Zhao, Y., Siriwardane, E. M. D., Wu, Z., Fu, N., Al- Fahdi, M., Hu, M., and Hu, J. Physics guided deep learning for generative design of crystal materials with symmetry constraints. npj Computational Materials, 9 (1):38, Mar 2023. ISSN 2057-3960. doi: 10.1038/s41524-023-00987-9. URL https://doi.org/10.1038/s41524-023-00987-9

About

Files and data needed to reproduce the machine learning results of the paper "Materials discovery using generative machine learning"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published