What's the difference between explain_message and edge_weight?

[ZechenGao]

I'm new to pytorch_geometric explain module. I found def explain_message(in message_passing.py) multiplies self._edge_mask and the out after message propagating. And edge_weight seems like dealing with the same way(eg. def message in gcn_conv.py). What's the diffrence between them? Thanks!

[kano5266]

Hi there!
The edge_mask in expain_message doesn't change the calculation of the node degree $\hat{d}_i = 1 + \sum_{j \in \mathcal{N}(i)} e_{j,i}$ in GCNConv. However, the edge weight does affect it.

Also, edge_mask can be used on all massage passing neural networks. While only a few MPNNs accept edge_weight as input.

Hope that helps!

[ZechenGao]

Thanks! That helps me a lot!