MLatticeABC: Machine learning for crystal lattice constant (a,b,c) prediction

Premise

MLatticeABC is a random forest machine learning model with a new feature set combined with the standard composition features such as Magpie descriptors for effective lattice constant prediction.

If you find our software is useful, please cite it as:

Yuxin Li, Wenhui Yang, Rongzhi Dong, Jianjun Hu*, MLatticeABC: Generic Lattice Constant Prediction of Inorganic Materials using Machine Learning, Arxiv. 2020

Developed in 2020.9-10 at
School of Mechanical Engineering
Guizhou University, Guiyang, China

Machine Learning and Evolution Laboratory
Department of Computer Science and Engineering
University of South Carolina, Columbia, USA

Performance on Materials Project dataset

Our models are trained with the dataset of 'Element_atom_no_stat+Enhanced_magpie.csv' by useing the RF.py, and the dataset can be downloaded from here data.zip. It also includes the dataset named 'Enhanced_magpie.csv' for the baseline algorithm.

Prediction performance of MLatticeABC in terms of R2 score for a,b,c over different crystal systems （10 repeats of random split 9:1)

Crystal system	train set size	test set size	a	b	c
Cubic	16492	1833	0.979±0.016
Hexagonal	8318	925	0.892±0.019		0.760±0.061
Trigonal	9977	1109	0.843±0.025		0.705±0.065
Tetragonal	13188	1466	0.846±0.024		0.685±0.044
Orthorhombic	24120	2681	0.770±0.019	0.579±0.074	0.638±0.028
Monoclinic	26884	2988	0.520±0.021	0.507±0.016	0.489±0.029
Triclinic	13767	1530	0.800±0.022	0.771±0.014	0.650±0.056

Environment Setup

To use MLatticeABC you need to create an environment with the correct dependencies. Using Anaconda this can be accomplished with the following commands:

conda create --name mlatticeabc python=3.6
conda activate mlatticeabc
conda install --channel conda-forge pymatgen
pip install matminer
pip install scikit-learn==0.23.1

MLatticeABC Setup

Once you have setup an environment with the correct dependencies you can install mlatticeabc using the following commands:

conda activate mlatticeabc
git clone https://github.com/usccolumbia/MLatticeABC
cd MLatticeABC
pip install -e .

Pre-trained models are stored in google drive. Download the file model.zip from from the figshare. After downing the file, copy it to MLatticeABC and extract it. the Model folder should be in the MLatticeABC directory after the extraction is completed.

Example Use

In order to test your installation you can run the following example from your MLatticeABC directory:

cd /path/to/MLatticeABC/
python predict.py -i full_formula -s crystal_system

for example:
python predict.py -i Mn16Zn24Ge24O96 -s cubic
python predict.py -i Mn16Zn24Ge24O96

The following cyrstal_system values are accepted

crystal     # crystal system unknown.
cubic
hexagonal
trigonal
tetragonal
orthorhombic
monoclinic
triclinic