PyTorch implementation of DeepMind Graph Nets. The original code depends on Tensorflow and Sonnet.
This implementation is based on PyTorch Geometric which is a geometric deep learning extension library for PyTorch
Graph Networks are a general framework that generalizes graph neural networks. It unifies Message Passing Neural Networks (MPNNs) and Non-Local Neural Networks (NLNNs), as well as other variants like Interaction Networks (INs) orRelation Networks (RNs).
You can have a look at Graph Networks in their arXiV paper: Battaglia, Peter W., et al. "Relational inductive biases, deep learning, and graph networks." arXiv preprint arXiv:1806.01261 (2018)
The following models are available:
- Interaction Network
- Graph Independent
You can also build your own models using the Blocks:
- Node Model
- Edge Model
PyTorch 1.8.0 and PyTorch Geometric.
We provide an example that tests the output against DeepMind's graph_nets.