How to convert a matrix to edge attribute and edge index

[riskiem]

I have a matrix as follows:

array([[0, 7, 4, 7],
       [7, 0, 9, 4],
       [4, 2, 0, 7],
       [7, 5, 8, 0]])

I want to convert this to the edge index and edge attribute for GNN. So for example for edge [1,2] I want the attribute to be 7. I am not sure how to go about this in pyotrch.

[wsad1]

One approach would be to use torch.nonzero

adj = torch.tensor([[0, 7, 4, 7],
       [7, 0, 9, 4],
       [4, 2, 0, 7],
       [7, 5, 8, 0]])
edge_index = torch.nonzero(adj,)
edge_attr = adj[torch.nonzero(adj, as_tuple= True)]

[rusty1s]

There also exists torch_geometric.utils.dense_to_sparse.