GCNConv error on hetero graph

[1AngelUS]

Hi,

I am implementing the following code from this https://pytorch-geometric.readthedocs.io/en/latest/notes/heterogeneous.html link.
I just replaced the GATConv layer with GCNConv layer.

class HeteroGNN(torch.nn.Module):
    def __init__(self, hidden_channels, out_channels, num_layers):
        super().__init__()

        self.convs = torch.nn.ModuleList()
        for _ in range(num_layers):
            conv = HeteroConv({
                ('paper', 'cites', 'paper'): GCNConv(-1, hidden_channels),
                ('paper', 'rev_writes', 'author'): GATConv((-1, -1), hidden_channels),
            }, aggr='sum')
            self.convs.append(conv)

        self.lin = Linear(hidden_channels, out_channels)

    def forward(self, x_dict, edge_index_dict):
        for conv in self.convs:
            x_dict = conv(x_dict, edge_index_dict)
            x_dict = {key: x.relu() for key, x in x_dict.items()}
        return self.lin(x_dict['author'])

model = HeteroGNN(hidden_channels=64, out_channels=dataset.num_classes,
                  num_layers=2)

It is however throwing the following error:
AttributeError: 'tuple' object has no attribute 'size' in 1st GCNConv layer. Could you please help me why it is not working.

Thank you!

[EdisonLeeeee]

Currently GCNConv is not supported for heterogeneous graphs. See here for the supported layers on heterogeneous graphs.

[1AngelUS]

Thank you for your quick reply!
I didn't know this. Thanks!

They have used GCNConv in this link also, so, I got confused.
https://pytorch-geometric.readthedocs.io/en/latest/notes/heterogeneous.html

[EdisonLeeeee]

In that example, GCNConv is only used when the source node and destination node share the same type. Actually, the relation src == dst is treated as a homogeneous one in HeteroConv:

pytorch_geometric/torch_geometric/nn/conv/hetero_conv.py

Lines 127 to 131 in a6b92fd

	if src == dst:
	out = conv(x_dict[src], edge_index, *args, **kwargs)
	else:
	out = conv((x_dict[src], x_dict[dst]), edge_index, *args,
	**kwargs)

[1AngelUS]

Sorry for the late reply.
Thank you for explaining it so well. I understand now :)
I will change GCNConv with some other GNN.