This was made following the Challenge 2 "Predict elasticity of Crystals"

What it does

In this project I did some predictions for the Stiffness Matrices and Piezoelectric Tensors of compounds in the Materials Project database.

How we built it

There are three Colab Notebook, two for predicting the Stiffness Tensor using Crystal Graph Neural Networks and the other using Random Forest, and the third Colab is for the prediction of the Piezoelectric Tensor using GCNN. For this I used pymatgen to retrieve the data from the Materials Project and DeepChem for the CGNN modules. In order to define features, for CGNN I used the structure data for the primitive cell as is given by pymatgen. On the other hand, the features for Random Forest where just the compound formulas, featurized in bot cases with DeepChem utilities.

Challenges we ran into

To learn how CGNN works, and to use the Materials Project dabatabase, which I have never used before.

Accomplishments that we're proud of

To make the CGNN actually work, as at first it was not working at all.

What we learned

How to define a dataset for use with DeepChem, and besides that, all the steps implied with pymatgen.

What's next for Prediction of Elasticity of Crystals

Continue exploring CGNN, as in general give better predictions.

Built With

Share this project: