Why do we need two loop in aggregate?

[lihuiliullh]

[liujian19911023]

Hi, do you figure out this?

[doujiang-zheng]

You can refer to the original algorithm in the paper. It doesn't mean a simple message passing from k-hop neighbors, but a l-layer transformation of all nodes (without the consideration of scalability). In each layer, all nodes are convolved with their neighbors.

[mxydls]

I don't understand either... Anyone who knows please help.

[doujiang-zheng]

I think this implementation is too obsecure to understand. I would recommend you refer to DGL GraphSAGE or PyG GraphSAGE, which are easier to use.

[mxydls]

I think this implementation is too obsecure to understand. I would recommend you refer to DGL GraphSAGE or PyG GraphSAGE, which are easier to use.

That's helpful. Thanks a lot!