Machine Learning for Property Prediction of Materials

Overview

This is a set of Python codes based on the Scikit-learn library to predict a property for a set of given materials. Our choice of materials is functionalized MXenes, and the property of interest is their GW-level band gap.

Prerequisites

Before using this package, ensure the following software/files is installed/available:

• Python
• Python Libraries :
  • Numpy
  • Scipy
  • matplotlib
  • scikit-learn (library for machine learning purposes)
• Jupyter Notebook

What will the script do

• Reads X (given features) and y (desired property)
• Generates primary and compound features (by applying math operations)
• Uses LASSO and Random Forest regressor to reduce/select important features
• Apply techniques such as linear regression and kernel ridge regression for property prediction

How to run the script?

Assuming that you have the necessary input files (features and the desired property for several materials),

you can simply run:

python ml-2nd-level.py

or

jupyter-notebook KRR-tune-hyperparam.ipynb

Additional Files and Regression Attempts

The notebook GBR-85x12-100_50_0.1.ipynb also has examples on

• SupportVectorMachine Regressor with GridSearchCV
• AdaBoostRegressor
• GradientBoostingRegressor

We can run this notebook by

jupyter-notebook GBR-85x12-100_50_0.1.ipynb

More information

More details about machine learning based property prediction can be found in our paper: A. C. Rajan et al. Chem. Mater. 2018.

License

© Arun Rajan @2018