Multiple-graph classification for graphs with different graph sizes



I am new to GCN but found DGL and its tutorial to be extremely useful.
In the DGL tutorial of Batched Graph Classification, each of the graph is of size [10, 20]. It is possible to use DGL on graph classification where each graph might be of different sizes?

If so, could you give some hints on how to do that? One way I can think of is to pad the graph adjacent matrix to the same size. Is there any better way for doing so?

Thank you.


Ah, I think I have figured out how to do this. DGL is quite handy actually. I think I can create a list of graphs where each graph could be of different sizes. Then I supply the list of graphs to dgl.batch() and follow the tutorial.


Indeed. Just go ahead :grinning:.