Question about adj values in cluster.py

[Antoine-ls]

I using ClusterGCN for node classification, and I want to get degree of one node in the adjacent matrix, however, I find that the adj is defined in cluster.py as follows, I do not understand why you set value to torch.arange(E, device=data.edge_index.device) instead of torch.ones(E), I want to calculate degree by adj.sum(0) but I can not do this in this adj, I wonder if this definition is necessary or if you have any suggestions for me to calculate degree in this sparse adj?

adj = SparseTensor(
row=data.edge_index[0], col=data.edge_index[1],
value=torch.arange(E, device=data.edge_index.device),
sparse_sizes=(N, N))

[rusty1s]

adj is just an internal storage of the ClusterData, which means that you probably shouldn't operate on it. We save a torch.arange vector in there as value in order to determine the original ID of edges when creating subgraphs (in order to filter edge features).

If you want to get the node degree of nodes in your graph, you can utilize torch_geometric.utils.degree, either on the original edge_index, or on the subgraph data data.edge_index returned by the ClusterData/ClusterLoader.

[Antoine-ls]

Understood, thanks for your reply!