edge_weight in GAT

[iknoorjobs]

Hi,

I see edge_weight can be used in GCN. Please tell how to use edge_weight in GAT?

Here (link) it's mentioned that "Note that the edge_weight will not be used."

Thanks

[iknoorjobs]

Also, please tell if we can have multi-dimensional edge weights. For example, using 2-dimensional edge_weight in the GCN example, it gives an error.

x = np.random.randn(5, 20).astype(np.float32)  # 5 nodes, 20 features
edge_index = np.array([
    [0, 0, 1, 3],
    [1, 2, 2, 1]
])
edge_weight = np.array([[0.9,0.5], [0.8,0.5], [0.1,0.5], [0.2,0.5]]).astype(np.float32)

Thanks

[hujunxianligong]

1. the edge_weight is not used by the GAT algorithm.
2. If you want to custom a special edge_weight, you can just custom your tfg GNN model and use it as a common input argument of the call method.