14fb6e1972270849d06f9eaad7aa9ec158d6308f,torch_geometric/nn/conv/pna_conv.py,PNAConv,forward,#PNAConv#Any#Any#Any#,223

Before Change

        self.mixing_network = FCLayer(self.out_channels, self.out_channels, activation=activation.LeakyReLU())

    def forward(self, x, edge_index, edge_attr):
        return self.propagate(edge_index, x=x, edge_attr=self.edge_encoder(edge_attr) if self.edge_features else None)

    def message(self, x_i, x_j, edge_attr):
        if self.divide_input:
            // divide the features among the towers

After Change

            x = x.view(-1, 1, self.F_in).repeat(1, self.towers, 1)

        // propagate_type: (x: Tensor, edge_attr: OptTensor)
        out = self.propagate(edge_index, x=x, edge_attr=edge_attr, size=None)

        out = torch.cat([x, out], dim=-1)
        outs = [nn(out[:, i]) for i, nn in enumerate(self.post_nns)]
        out = torch.cat(outs, dim=1)

        return self.lin(out)

    def message(self, x_i: Tensor, x_j: Tensor,
                edge_attr: OptTensor) -> Tensor:

Instances