ai4materials allows to perform complex analysis of materials science data using machine learning. It also provide functions to pre-process (on parallel processors), save and subsequently load materials science datasets, thus easing the traceability, reproducibility, and prototyping of new models.
An online documentation can be found here: https://ai4materials.readthedocs.io/en/latest/
Code authors: Angelo Ziletti, Ph.D. ([email protected]), Andreas Leitherer ([email protected], [email protected])
========================================================
ai4materials provides code for:
@article{Leitherer2021,
doi = {10.1038/s41467-021-26511-5},
url = {https://doi.org/10.1038/s41467-021-26511-5},
year = {2021},
month = oct,
publisher = {Springer Science and Business Media {LLC}},
volume = {12},
number = {1},
author = {Andreas Leitherer and Angelo Ziletti and Luca M. Ghiringhelli},
title = {Robust recognition and exploratory analysis of crystal structures via Bayesian deep learning},
journal = {Nature Communications}
}
Please cite this work if you use ai4materials in your work or research.
You can proceed with the installation steps as described below or directly proceed to a tutorial available at
http://analytics-toolkit.nomad-coe.eu/tutorial-ARISE
within the NOMAD analytics toolkit (https://nomad-lab.eu/AItutorials) where you do not have to install any software.
The code is currently under active development.
We recommend to create a virtual python 3.7 environment (for instance, with conda: https://docs.anaconda.com/anaconda/install/linux/), and then execute
pip install 'git+https://github.com/angeloziletti/ai4materials.git'
To utilize the deep learning techniques introduced in A. Leitherer, A. Ziletti, and L.M. Ghiringhelli, 2021 (https://arxiv.org/abs/2103.09777), you need to install the quippy package (https://github.com/libAtoms/QUIP). This is necessary to compute the SOAP descriptor.
The following commands to install QUIP have been tested for a python 3.7 environment created using conda
Note that when executing make config, choose always the standard options, i.e., simply hit enter, execept when asked for compiling with GAP support, enter 'y'. Please check also for additional dependencies you may need to install, see also https://libatoms.github.io/GAP/installation.html):
git clone --recursive https://github.com/libAtoms/QUIP.git
cd QUIP/
export QUIP_ARCH=linux_x86_64_gfortran
make config
make
git submodule update --init --recursive
pip install f90wrap
make install-quippy
For global or local analysis of single- or polycrystalline systems, one just needs to define the corresponding geometry file and load a pretrained model for prediction:
from ai4materials.models import ARISE
geometry_files = [ file_1, file_2, ... ]
predictions, uncertainty = ARISE.analyze(geometry_files, mode='global')
predictions, uncertainty = ARISE.analyze(geometry_files, mode='local',
stride=[[4.0, 4.0, 4.0], ...], box_size=[12.0, ...])
Please refer to http://analytics-toolkit.nomad-coe.eu/tutorial-ARISE and the associated publication for more details.