How should I do withnegative_sampling() for bipartite graph

[ogawayuto]

HI, I have a question about torch_geometric.utils.negative_sampling(). In the refrence you can give (num_src_node, num_dst_node) as num_nodes argument for bipartite graph. In the refrence you can see an example below.
(https://pytorch-geometric.readthedocs.io/en/latest/modules/utils.html#torch_geometric.utils.negative_sampling)

edge_index = torch.as_tensor([[0, 0, 1, 2],
                             [0, 1, 2, 3]])
negative_sampling(edge_index, num_nodes=(3, 4))
 >>> tensor([[0, 2, 2, 1],
        [2, 2, 1, 3]])

First, I assumed a graph with 3 nodes and a graph with 4 nodes and the function would return negative edges between two graphs. But, the example seems to return edges edges like self-loop, internal edges, edges between graphs.

I'm confused with what the funcions do when you give bipartite graph. How should I do to get negative edges between two types of nodes in bipartite graph?

[akihironitta]

the example seems to return edges edges like self-loop, internal edges, edges between graphs.

I believe the example is correct. It looks like a self-loop, but it's an edge from 2 (src node type) to 2 (dst node type).

[ogawayuto]

Thank you for your reply! I was confusing src indices and dst indices.