GCNConv for batch

[shlee-home]

Hi, thank you for opening these codes.
I'm progressing a study using graph neural network. However, I wonder that how can I use GCNConv for training when I train the model using batch?

When I run following code:

class GConv(nn.Module):
    def __init__(self, input_dims, output_dims, hidden_dims, num_layers):
        super(GConv, self).__init__()
        self.layers = nn.ModuleList()
        self.batch_norms = nn.ModuleList()

        for i in range(num_layers):
            if i == 0:
                self.layers.append(GCNConv(input_dims, hidden_dims))
            else:
                self.layers.append(GCNConv(hidden_dims, hidden_dims))
            self.batch_norms.append(nn.BatchNorm1d(hidden_dims))

        project_dims = hidden_dims * num_layers
        self.project = torch.nn.Sequential(
            nn.Linear(project_dims, output_dims),
            nn.ReLU(inplace=True),
            nn.Linear(output_dims, output_dims),
            nn.ReLU(inplace=True))

    def forward(self, x, edge_index, edge_weight, batch):
        z = x
        zs = []
        for conv, bn in zip(self.layers, self.batch_norms):
            print(z.size(), edge_index.size(), edge_weight.view(-1, 1).size())
            z = conv(z, edge_index, edge_weight.to(torch.float32).view(-1, 1))
            print(z.shape)
            z = F.relu(z)
            z = bn(z)
            zs.append(z)
        gs = [global_add_pool(z, batch) for z in zs]
        z, g = [torch.cat(x, dim=1) for x in [zs, gs]]
        return z, g

class Encoder(torch.nn.Module):
    def __init__(self, encoder, augmentor):
        super(Encoder, self).__init__()
        self.encoder = encoder

    def forward(self, x, edge_index, edge_weight, batch):
        print(x.size(), edge_index.size(), edge_weight.size(), batch.size())
        z, g = self.encoder(x, edge_index, edge_weight, batch)`

It returns an error:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (3000*1 and 300 * 64)

I think the mat1's shape (batch_size*num_nodes, node_feature_dims) and mat2's shape (num_nodes, hidden_dims)
Any help would be greatly appreciated :)

[rusty1s]

In which function does this error occur? What are the shapes of its inputs of:

print(x.size(), edge_index.size(), edge_weight.size(), batch.size())

The implementation looks correct to me, so I assume there must be some inconsistencies in the input arguments.