Code and data associated with the publication "Machine-learning improves understanding of glass formation in metallic systems".

https://doi.org/10.1039/D2DD00026A

The code in this repository utilises a number of other packages to process data, train neural networks, evaluate those networks, and visualise predictions.

To run this code, execute the following:

git clone https://github.com/Robert-Forrest/GFA
cd GFA
python3 -m pip install -r requirements.txt
python3 __main__.py examples/simple.yaml

The examples directory contains configuration files for a number of situations.

simple.yaml contains configuration for the simple task, which trains a standard neural-network model. The prediction targets are defined in the targets list.

kfolds.yaml contains configuration for the kfolds task, which performs k-folds cross-validation on the standard neural-network model.

kfoldsEnsemble.yaml contains configuration for the kfoldsEnsemble task, which performs ensembling to create a meta-learner based on the submodels produced during k-folds cross-validation.

permutation.yaml contains configuration for the feature_permutation task, which shuffles features and measures the resulting change in model efficacy to judge their importance.

composition_scan.yaml contains configuration for the composition_scan task, which takes as input alloy spaces such as CuZr or FeNiBe, and creates graphs of features and predictions across composition space.