How to See Hidden Patterns in Metamaterials with Interpretable Machine Learning

This repository contains the code for experiments in the following paper

Zhi Chen, Alexander Ogren, Chiara Daraio, L. Catherine Brinson, Cynthia Rudin

How to See Hidden Patterns in Metamaterials with Interpretable Machine Learning

Extreme Mechanics Letters. https://doi.org/10.1016/j.eml.2022.101895

arXiv pre-publication version

Data

Please download bandgap_data.mat, containing the 10 by 10 metamaterials unit-cells and their simulated (by FEA) dispersion curves, from this google drive folder. After the dataset is downloaded, please put them in the /data folder

Code Details

Structure-to-property Prediction

Run

python3 compare_bacc.py

to compare the test balanced-accuracies of different ML models trained on raw features and shape-frequency features for predicting bandgap existence.

Train sparse decision trees on shape frequency features

Run

python3 create_bin_datasets.py

to create binarized datasets to be used by GOSDT. Then run

python3 run_gosdt_sff.py

to get the sparse decision trees trained via GOSDT algorithm.

Train Unit-cell Template Sets

The unit-cell template algorithm contains two steps (1) preselection; (2) MIP optimization Run

python3 preselect_templates.py

to preselect a useful set of templates. Then run

python3 choose_templates.py

to get the optimal set of templates using MIP. Note that, we use CPLEX optimizer to solve the MIP, please install it before running choose_templates.py.