This directory contains pre-trained models from the Materials Virtual Lab.
Currently, we provide models for
- QM9 molecules
- Materials Project (MP) crystals
- Multi-fidelity band gap data
The original models in "Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals". are in the folders dated 2018.6.1. We are continuing to develop new models on updated datasets, and these are dated accordingly (e.g., mp-2019.4.1 contains models developed on the much larger Materials Project dataset of 133,420 crystals as of Apr 1 2019).
The README.md in each folder provides a summary of the dataset, training/test split and the model performance.