Why clamp in DenseSageConv?

[mdanb]

In the forward method for DenseSAGEConv, there is the line:
out = out / adj.sum(dim=-1, keepdim=True).clamp(min=1)
I understand that clamping will prevent division by 0 if the degree of a node is 0, but what if the edges are weighted, in which case the "degree" can be between 0 and 1?

[rusty1s]

Adjacency values in DenseSAGEConv should not be weighted, and one should use DenseGraphConv instead. I need to add this to the documentation.