GCN Implementation in the form of MessagPasing with EdgeFeature

[pouya-taghipour]

Hi there,

Implementation of GCN algorithm in the fom of MessagPassing is described in PytorchGeometric MessagePassin class documentation (see the https://pytorch-geometric.readthedocs.io/en/latest/tutorial/create_gnn.html)

However, I also passe the edge feature in propagate function in my implementation as below:

def forward(self, x, edge_index, edge_attr):
        x = self.linear(x)
        edge_embedding = self.edge_encoder(edge_attr)

        row, col = edge_index

        deg = degree(row, x.size(0), dtype = x.dtype) + 1
        deg_inv_sqrt = deg.pow(-0.5)
        deg_inv_sqrt[deg_inv_sqrt == float('inf')] = 0
        norm = deg_inv_sqrt[row] * deg_inv_sqrt[col]

        return self.propagate(edge_index, x=x, edge_attr = edge_embedding, norm=norm)

My question is how propagate function incorporate edge feture? Does propagate function sum the edge and node feature of neighbors as below?

is the edge feature

or

we will have seperate weight for node and edge as below: