How to integrate multi-dimensional edge features to GCN?

[Shiiivaa]

Hello,
Currently, I am using a combination of GCN and GAT layers for a node regression task in my GNN structure. This setup works well when using only node features. However, I now want to incorporate multi-dimensional edge features into my GNN structure.
The problem is that GCN seems to only accept 1-dimensional features (edge weights). Can you help me figure out how to integrate multi-dimensional edge features into GCN while still maintaining the current structure (i.e., combination of GCN and GAT layers)? I would prefer not to switch to a different GNN structure.

[wsad1]

1. I think its best to not use a GCN if you want to use multi-dimensional edge features. Instead you could use any layer that supports edge_attr from here, like GATConv.
2. If you really want to stick with GCN you could apply a linear transform to your edge_attr to transform it to a 1-d tensor and pass it as edge_wieght to GCN.

[riyajatar37003]

so let's assume if i have edge feature are in 2dim. u mean to say first project them into 1-d and use them as edge weight.?
should i keep this linear transform layer as learnable.?