SAGEConv('gcn') vs GraphConv(default)

Questions
#1

Is
SAGEConv(aggregator_type=“gcn”)
totally equivalent to the default GraphConv:

GraphConv(norm=‘both’, weight=True, bias=True, activation=None, allow_zero_in_degree=False)?

If no, could you please suggest what script i should add/modify? (eg the norm part?)

Thanks a lot!

#2

They are not the same. “GCN” aggregator is referring to the “convolutional” variant, mentioned in the context around Eq 2 in the original paper. In particular, their normalization constants are different.

1 Like