Thank the DGL team for creating such a great tool.
I have a question about the way to get dis_inputs, specifically this line:
v[:g.number_of_dst_nodes(k)]
is subseting the first n entries of the input feature matrix, where n=g.number_of_dst_nodes(k) being the number of dst nodes. Unless the input feature matrix is ordered in a way that the first n entries corresponding to dst nodes, I don’t really understand why we could simply subset. Are we need to do something like v[dst_nids]?
Similar logics also applies to specific layers like https://github.com/dmlc/dgl/blob/8bbc84e04fe33a5459b85ad09e92b5a78325f214/python/dgl/nn/pytorch/conv/sageconv.py#L181
I read through the code but could not figure out how. Looking forward to your answer.