search

pyg-team / pytorch_geometric

Graph Neural Network Library for PyTorch
https://pyg.org
MIT License
21.09k stars 3.63k forks source link

More operations over heterogeneous graphs #1907

Open douglasrizzo opened 3 years ago

douglasrizzo commented 3 years ago

🚀 Feature

Recently, a survey has been made available online, listing multiple machine learning techniques specialized in data structured as heterogeneous graphs (HGs). Some of them may already be present in PyG (I've personally used RGCN myself), but the survey can be used for future reference when considering new operations to be added to PyG.

Motivation

There are lots of applications which involve HGs. I remember when I first used PyG in the beginning of 2020 and the only recommendation I got to work with graphs in which nodes have types was RGCN. This survey has done a good job in finding and categorizing other papers that focus on machine learning applied to HGs. They also have gathered links to public implementations of some of the papers. I believe the linked document is a good reference for selecting future graph operations to implement in PyG.

rusty1s commented 3 years ago

Thanks for the pointer. I think we are on good track regarding heterogeneous graph support in PyG, e.g. via RGCN, bipartite GNN ops, metapath2vec, but this document helps a lot :)