Evolutionary design of machine-learning-predicted bulk metallic glasses

Code and data associated with the publication "Evolutionary design of machine-learning-predicted bulk metallic glasses".

doi:xyz

Instructions

The code in this repository utilises a number of other packages to run genetic algorithms, evaluate machine-learning models, and visualise predictions. For optimisation of properties discussed in the paper, a model trained to predict those properties must be available - such a model can be created using techniques discussed in our previous publication (https://doi.org/10.1039/D2DD00026A) using the cerebral package. The evomatic package is used here to perform genetic algorithm searching.

To prepare to run this code, execute the following:

git clone https://github.com/Robert-Forrest/EvolutionaryDesignOfBMGs
cd EvolutionaryDesignOfBMGs
python3 -m pip install -r requirements.txt

There are various YAML configuration file examples provided demonstrating searching scenarios with different targets and constraints. The example congfiguration for a simple search for high-Dmax alloy compositions may be invoked as follows:

python3 __main__.py Dmax.yaml

Custom YAML files may easily be created and invoked in a similar way.

Running main.py will perform the search as configured, and will output a variety of figures and data-files to the output directory, describing the alloy population it observed.

To enable calculations of the novelty of proposed alloy compositions, provide a list of data filepaths via the `model_datafile' parameter in the configuration file.