/pytorch-graphsage

Representation learning on large graphs using stochastic graph convolutions.

Primary LanguagePython

graphsage

pytorch version of https://github.com/williamleif/GraphSAGE

There are also a handful of new features, including:

  • scripts for preprocessing data
  • attention-based aggregator
  • sparse edge sampler (eg, don't use the dense 2D edgelist)
  • richer, pluggable preprocessing classes

This version is missing some features, as well. Namely:

  • we don't support unsupervised models (yet)

A simplified version of the original Tensorflow version can be found on branch tf-simple.

LICENSE

MIT