GraphSage Convolution on nodes with multiple self-loops?

Hello guys,
I have mistakenly added two self loops to a set of nodes and I performed GraphSage convolution on the graph. Will its previous state representations be included two times in the aggregation step ?

each neighbor will be aggregated according to SAGEConv — DGL 1.1 documentation. Namely, self-looped twice node will be aggregated twice. But this may not change the output of conv(g, feat) due to aggregate type such as pool.

1 Like

@Rhett-Ying Makes sense, if I do understand correctly the max pooling is what would make the final output unchanged. In my case its a mean aggregation so that would definitely affect the output.