Multiple reduce functions

Questions
#1

Hi!

I’d like to use different reduce functions based on the node type.
For example, if the node’s type is A, then sum incoming vectors from edges and multiply it with matrix W_a, and if the node’s type is B, then multiply sum with matrix W_b.
So, can you give me some advice, what is an efficient way to do this?
I know, that I can iterate over NodeBatch and choose this matrix individually, but this will kill all batching…

#2

Hi @SpirinEgor, I think your use case is the same as RGCN’s, DGL supports heterogeneous graph where you can do multi-type message passing synchronously.

Please refer the heterogeneous graph tutorial and see how RGCN is implemented:
https://docs.dgl.ai/tutorials/hetero/1_basics.html#relational-gcn-on-heterograph