Machine learning for solid solutions

This document provides instructions for running machine learning models downloaded from GitHub repository.

• The necessary conda environment can be imported from the file 'environment.yml'.

On Linux/MacOS:

conda env create -f environment.yml
conda activate ml-solid-solns

1.1 Data extraction

We provide the raw data (i.e. DFT outputs) with notebooks for pre-processing them. This pre-processing can take some time so we also provide pre-processed data sets ready to be used with the different fitting procedures.

These instructions are for extracting vasp energies and Coulomb matrix, prior to machine learning.

It is assumed that the default directory structure, as downloaded from the repository, remains in place.

• 'repository-data.zip' should be downloaded from Zenodo (https://doi.org/10.5281/zenodo.4736810) and unzippled in the top level of this repo.
• You can use pre-generated files if you wish, this is found in repository-data/coulomb_matrix/input_data_cme.pkl
• If you want to do your own pre-processing; data.ipynb notebook can be found in coulomb_matrix/data/rundir/ this can be used to generate the file required.

These instructions are for extracting vasp energies and cluster correlation functions, prior to machine learning.

• 'repository-data.zip' should be downloaded from Zenodo (URL: ) and unzipped in the top level of this repo.
• You can use pre-generated files if you wish, this is found in repository-data/cluster_correlation_functions/input_data_ccf.pkl
• If you wan to do your own pre-processing; data-ccf.ipynb notebook can be found in cluster_correlation_functions/data/rundir/ this can be used to generate the file required.

1.2. Machine learning

Machine learning notebooks ('mlp.ipynb', 'gbdt.ipynb', 'lr-emix.ipynb' and 'lr-egap.ipynb') can be found in sub-directories.

• N.B. Users should specify the desired percentage of data to be used for training, by manually editing the second cell of the notebook, prior to training the model.
• There are various parts of the code which may be edited by users, however the two main user inputs are located in the first two cells of each notebook.