Graph Convolutional Neural Networks

Unofficial Reimplementation of "Semi-Supervised Classification with Graph Convolutional Networks"¹² in PyTorch.

Usage

Dense

input= ...
adj= ...
n_nodes=adj.size(0) # or n_nodes = input.size(1)n_features=input.size(-1)
h1_features= ...
h2_features= ...
conv1=nn.Sequential(LinearGraphConv(n_features, h1_features), nn.ReLU())
conv2=nn.Sequential(LinearGraphConv(h1_features, h2_features), nn.ReLU())
output=conv2((conv1((input, adj)), adj))

Sparse

input= ...
adj_sparse_coo= ... n_nodes=adj.size(0) # or n_nodes = input.size(1)n_features=input.size(-1)
h1_features= ...
h2_features= ...
conv1=nn.Sequential(SparseLinearGraphConv(n_features, h1_features), nn.ReLU())
conv2=nn.Sequential(SparseLinearGraphConv(h1_features, h2_features), nn.ReLU())
output=conv2((conv1((input, adj_sparse_coo)), adj_sparse_coo))

TODOs

Add Cora dataset example.

References

Footnotes

1. Kipf & Welling, Semi-Supervised Classification with Graph Convolutional Networks, 2016. ↩

2. Official Implementation. ↩