How to impement a fully adjacent layer in GCN?

[codeastra2]

I would like to implement a fully adjacent layer in GNN models. i currently have an architecture like so:

        self.conv1 = GCNConv(num_node_features, 16)
        self.conv2 = GCNConv(16, 16)
        self.conv3 = GCNConv(16, 16)
        self.conv4 = GCNConv(16, 16)

How can I make conv4 have a fully adjacent network?

[rusty1s]

You mean a fully-connected graph? That is determined by the input, i.e., edge_index. You can generate a fully connected graph via

edge_index = torch.ones(num_nodes, num_nodes).nonzero().t()

but keep in mind that this may be slow to run since we are basically using an operator designed for sparse inputs on a fully-connected graph.