TypeError: forward() takes 2 positional arguments but 3 were given

[xyqbest]

I want to use residual units in my GCN net. However, I meet the argument number error.
The error logs are

Traceback (most recent call last):
  File "D:\xxx\pythonProject\phdThesis\del_debug.py", line 153, in <module>
    run_GCN(m_GCN, res_path="./results/gcn_xpe_u_u", is_weight=False)
  File "D:\xxx\pythonProject\phdThesis\del_debug.py", line 32, in run_GCN
    out = model(x=data.x, edge_index=data.edge_index)
  File "D:\ProgramData\Anaconda3\envs\phdThesis\lib\site-packages\torch\nn\modules\module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "D:\xxx\pythonProject\phdThesis\del_debug.py", line 128, in forward
    x = self.ResNet(x, edge_index)
  File "D:\ProgramData\Anaconda3\envs\phdThesis\lib\site-packages\torch\nn\modules\module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
TypeError: forward() takes 2 positional arguments but 3 were given

line 153 is run_GCN(m_GCN, res_path="./results/gcn_xpe_u_u", is_weight=False)
line 32 is out = model(x=data.x, edge_index=data.edge_index)
line 128 is x = self.ResNet(x, edge_index)

It means that ResNet.forward receive 3 arguments ? but I just put 2 arguments (x and edge_index).

Here is my GCN network:

class myGCN(torch.nn.Module):
    def __init__(self, x_dim=8, output_dim=2, is_res=False, num_res_units=3):
        self.x_dim, self.output_dim = x_dim, output_dim
        self.is_res = is_res
        super(myGCN, self).__init__()
        self.num_res_units = num_res_units
        self.relu = nn.ReLU()
        self.bn1 = nn.BatchNorm1d(self.x_dim)
        self.bn2 = nn.BatchNorm1d(ceil(2 * self.x_dim))
        self.gcnInput = GCNConv(self.x_dim, ceil(2 * self.x_dim))
        self.gcnOutput = GCNConv(ceil(2 * self.x_dim), self.output_dim)
        self.nets = [ResUnitGCN(ceil(2 * self.x_dim), ceil(2 * self.x_dim), self.is_res)
                     for _ in range(self.num_res_units)]
        self.ResNet = nn.Sequential(*self.nets)

    def forward(self, x, edge_index=None, edge_weight=None):
        if edge_weight is not None:
            x = self.relu(self.gcnInput(self.bn1(x), edge_index, edge_weight))
            x = self.ResNet(x, edge_index, edge_weight)
            x = self.relu(self.gcnOutput(self.bn2(x), edge_index, edge_weight))
        else:
            x = self.relu(self.gcnInput(self.bn1(x), edge_index))
            x = self.ResNet(x, edge_index)
            x = self.relu(self.gcnOutput(self.bn2(x), edge_index))
        res = F.dropout(x, p=0.25)
        return res

Here is my residual unit:

class ResUnitGCN(nn.Module):
    def __init__(self, in_channels, out_channels, is_res=False):
        super(ResUnitGCN, self).__init__()
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.is_res = is_res

        self.relu = nn.ReLU(True)
        self.gcn1 = torch_geometric.nn.GCNConv(self.in_channels, self.out_channels)
        self.bn1 = nn.BatchNorm1d(self.out_channels)
        self.gcn2 = torch_geometric.nn.GCNConv(self.out_channels, self.out_channels)
        self.bn2 = nn.BatchNorm1d(self.out_channels)

    def forward(self, x, edge_index=None, edge_weight=None):
        if edge_weight is not None:
            x = self.relu(self.gcn1(self.bn1(x), edge_index, edge_weight))
            x = self.relu(self.gcn2(self.bn2(x), edge_index, edge_weight))
        else:
            x = self.relu(self.gcn1(self.bn1(x), edge_index))
            x = self.relu(self.gcn2(self.bn2(x), edge_index))
        out = x
        if self.is_res:
            out = out + x
        return out

How can I fix this error?
@rusty1s Could you help me? 😄

[rusty1s]

Yes, you sadly cannot use torch.nn.Sequential as part of a GNN pipeline. torch.nn.Sequential always expect two arguments (self and x), while GNN layers expect more (self, x and edge_index). As such, we have integrated our own Sequential implementation into PyG, see here.

[xyqbest]

Thank you. It works. 😄