keeeto / explainable-trees

An intro to training a decision tree model for materials band gap predicition and then using TreeExplainer to understand the model predictions.

Explainable materials machine learning

This is a set of notebooks intended to give a quick introductions into some methods for building and examining models that could be useful for materials design.

The first notebook classical-ml introduces a number of methods for fitting some features to data on the band gap of materials. The final model that we come to is based strongly on Data-Driven Discovery of Photoactive Quaternary Oxides Using First-Principles Machine Learning

The second notebook shapley_values_gbtree introduces the application of TreeExplainer to examine how the features of the model contribute to the outcomes. And to help with understanding the predictions that are made.

Files

• data - contains all the data needed to train the models
• models - contains a pre-trained decision tree, if you want to skip straight to tutorial 2
• notebooks - has the two notebooks
• environment.yml - contains the conda environment that these notebooks were developed in