Our second sprint, which took place in December focused on explainability. We had 18 contributors submitting over 30 commits and contributing to complex projects. You can find our roadmap for more details.

Thanks to the hard work of our community, PyG now supports:

A new support framework and unified interface to enable new explainability methods
Examples of the explainer methods using new framework
Adaptation of existing GNN explainability using the framework
Several benchmark data sets for evaluation
A broader set of evaluation metrics - including fidelity, AUC, faithfulness, utilities to combine fidelities for characterization scores, precision, recall, and more
Extensions to explainability methods for link prediction and heterogenous graphs
Better visualization tooling
Productionized research in new methods such as PGExplainer and SubgraphX
Demos, docs, tutorials, and more! See our explainer class as an example

We hit our goals and kicked off new collaborations to continue building off of the great work accomplished in the sprint. We're excited to continue expanding explainability into the next PyG release!

Related Articles:

Share on

You Might Also Like

Cookies settings