Error in `scatter_sum` with pre-transform that adds a node to the graph

❓ Questions & Help

Hi!

I get an error I really don’t understand in scatter_sum that only happens with a new transform that I made.

ERROR:

RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
  File "/Users/linda/miniconda2/envs/GraphNeuralNetworkExperiments/lib/python3.8/site-packages/torch_scatter/scatter.py", line 12, in scatter_sum
                out: Optional[torch.Tensor] = None,
                dim_size: Optional[int] = None) -> torch.Tensor:
    index = broadcast(index, src, dim)
            ~~~~~~~~~ <--- HERE
    if out is None:
        size = list(src.size())
  File "/Users/linda/miniconda2/envs/GraphNeuralNetworkExperiments/lib/python3.8/site-packages/torch_scatter/utils.py", line 13, in broadcast
    for _ in range(src.dim(), other.dim()):
        src = src.unsqueeze(-1)
    src = src.expand_as(other)
          ~~~~~~~~~~~~~ <--- HERE
    return src
RuntimeError: The expanded size of the tensor (212) must match the existing size (196) at non-singleton dimension 0.  Target sizes: [212, 128].  Tensor sizes: [196, 1]

PRE-TRANSFORM:

class CentralNode(object):
    def __call__(self, data):
        # add the central node
        central_node_features = torch.mean(data.x, 0)
        data.x = torch.cat((data.x, central_node_features.unsqueeze(0)), 0)

        # add edges to all other nodes
        additional_edges = torch.tensor([list(range(data.num_nodes)), [data.num_nodes]*data.num_nodes])
        data.edge_index = torch.cat((data.edge_index, additional_edges), 1)

        # give it a position
        if data.pos is not None:
            central_node_pos = torch.mean(data.pos, 0)
            data.pos = torch.cat((data.pos, central_node_pos.unsqueeze(0)), 0)

        return data

    def __repr__(self):
        return '{}()'.format(self.__class__.__name__)

Without the pre-transform, everything works fine. Any help is appreciated 😃! I

[rusty1s]

It looks like this may be a problem with your data. Please ensure that edge_index.max() is lower than x.size(0).

[rmunia]

Can you clarify? The assertion ensures that there exists node features for every index present in your edge_index. If your edge_index contains a value of 6742, that means you have at least 6742 nodes in your graph, thus x.size(0) should be of size 6742 as well.

Thank you for your explanation. For this, I could solve the issue.