Aggregation Issue

[Swknight]

Coagulation for your awesome work!
However, I have the question about the aggregation method in you code.
I found the features is concatenated to the result of the DNN and activation function in one_propagate function as the code:

features = self.mess_dropout(torch.cat([self.act(
                dnns[i](torch.matmul(graph, features))), features], 1))

But, according to the prior paper, they were always first concatenated the features and then pass the DNN and activation function, as:
features = self.mess_dropout(self.act(dnns[i](torch.cat([torch.matmul(graph, features), features], 1))))

If there are some special consideration? Look forward to your reply, thanks!