Integrating source_to_target and target_to_source flow in bipartite graph.

❓ Questions & Help

Hello! Thank you for building such a well-documented and well-maintained library. I’m new to Graph Neural Networks, and I am learning to build a bipartite graph network, for the following nodes:

I have gone through issues page to understand how to integrate source_to_target and target_to_source flows for some nodes. So currently, I made 2 BipartiteData objects, where edge_index is defined differently:

x_s = ... # size [7, 300]
x_t = ... # size [7, 11]
edge_index_1 = torch.tensor([[0, 1, 4, 4, 4, 4, 6], [3, 1, 0, 2, 5, 6, 4]], dtype=torch.long) #for source_to_target
edge_index_2 = torch.tensor([0, 1, 2, 3, 4, 5, 6], [0, 1, 2, 3, 4, 5, 6]], dtype=torch.long) #for target_to_source
graph_data_1 = BipartiteData(x_s=x_s, edge_index=edge_index_1, x_t=x_t)
graph_data_2 = BipartiteData(x_s=x_s, edge_index=edge_index_2, x_t=x_t)

Then, graph_data_1 and graph_data_2 are passed through separate GCN layers; where one overrides MessagePassing with flow = 'source_to_target' and the other overrides MessagePassing with flow = 'target_to_source'.

Is this a good approach to this problem? Also, how do I further join the outputs of the 2 GCN layers and get an overall node representation?

Your clarification will be appreciated 😃

[rusty1s]

Yeah, the batch object is not present because PyG cannot identify how to create it. You can pass the follow_batch argument to the DataLoader to signal PyG which batches it should track:

def __inc__(self, key, value):
      # Returns the incremental count to cumulatively increase the value
      # of the next attribute of :obj:`key` when creating batches.
      if (key == 'edge_index_s2t':
          return torch.tensor([[self.x_s.size(0)], [self.x_t.size(0)]])
      elif (key == 'edge_index_t2s':
          return torch.tensor([[self.x_t.size(0)], [self.x_s.size(0)]])
      else:
          return super(BipartiteData, self).__inc__(key, value)


loader = DataLoader(dataset, batch_size=..., follow_batch=['x_s', 'x_t'])

[rusty1s]

You can also define it as

class BipartiteData(Data):
    def __init__(self, edge_index_s2t, edge_idnex_t2s, x_s, x_t):
        super(BipartiteData, self).__init__()
        self.edge_index_s2t = edge_index_s2t
        self.edge_index_t2s = edge_index_t2s
        self.x_s = x_s
        self.x_t = x_t

    def __inc__(self, key, item):
        ...

There is no limitation in how many attributes you can store in data.