Can GCNConv layer handle adjacency matrices of weighted directed graphs that have negative weights?

[Laknath1996]

I am dealing with a dataset of weighted directed graphs and some of their adjacency matrices have negative entries. I am hoping to construct a simple GCN-based network for them, but when I attempt to train the network, the loss is "nan" from the very beginning. Therefore, I was wondering whether the adjacency matrices supported by the GCNConv layer allow negative entries within them. I'd be grateful if you could clarify my doubt.

[rusty1s]

Please seem my answer here: #397 (comment)