Weight Initialization

[gdemirezen]

Hello,
How can we initialize weights and biases of GNN operators like GraphConv, GATConv etc... in PyG?

[rusty1s]

Just as you would do in PyTorch. For example, if you want to initialize the weights of GCNConv, you can look up its code, see that is just uses a single Linear layer, and modify the weights of it, e.g.,

torch.nn.init.xavier_uniform(conv.lin.weight)

[gdemirezen]

Thank you for your quick reply!