Machine learning for entropy forming ability (EFA)

A lightweight training and evaluation framework for reliable determination of synthesizability (EFA) of high entropy materials.

See the associated publication for more information about EFA: https://www.nature.com/articles/s41524-020-0317-6

Setup

This project targets a CPU-enabled Linux workstation. Additional work may be required for testing on other operating systems.

We target Python 3.6+.

1. Create/Activate a virtual environment (via anaconda, virtualenv, or pyenv) Recommended: Anaconda
2. pip install -e .

How to Run Experiments

Train a model

This project includes a cli.py script for training a Random Forest Regressor with the following assumptions:

• All of your data is in an excel document.
• Your labels are the last column of the data file.
• The path to the data file is passed in as an argument.
• The output file names are currently generic plus a time stamp, but can easily be modified to contain more experimental details.

The cli.py script returns several artifacts including the fit model, a table of feature importances, the hyperparamter search results and optimal hyperparameters found, and a transform mask (if feature selection is performed).

Predicting EFA for new materials

This project includes a predict.py script for predicting the EFA of new data. The following assumptions are made:

• All of your data is in an excel document.
• The last column of the data file contains a header, but at least one of the cells in the column is blank. Celss with values will cause the material to be skipped.
• The path to the data file is passed in as an argument. -The path to the transform mask (if applicable) and the fit model must be updated within the script.
• The output file names are currently generic plus a time stamp, but can easily be modified to contain more experimental details.

The predict.py script returns a .csv file containing the material name and the predicted EFA.

Visualizing the decision trees

This project includes a predict.py script for exporting all of the decision trees in a random forest as png files.

This script assumes:

• The path to the model is updated

model_path = '../model_checkpoints/model_checkpoint_wCalphad_gs_2019-09-17-02-09.joblib'

• The data file is passed in as an argument.
• The transform mask (if applicable) is updated.

transform_mask_init = pd.read_csv('../transform_mask/FILENAME_HERE.csv')

Data provided in this repo:

• The 56 high entropy carbides and their EFA values from density functional theory (DFT).
• The 70 new Cr-containing carbides
• Pre-fit models using the 56 data points with DFT-based labels.
• Definitions of the predictor variable names Predictor-Variable-Definitions.pdf