Index out of Bounds for Link Prediction

[karan96]

Greetings,

I have been following this Link Prediction Notebook tutorial for my own custom dataset. The dataset looks like this: -

HeteroData(
  location={ node_id=[71] },
  experts={ node_id=[13631] },
  skills={ node_id=[69679] },
  (location, of, experts)={ edge_index=[2, 13631] },
  (skills, of, experts)={ edge_index=[2, 135690] },
  (experts, rev_of, location)={ edge_index=[2, 13631] },
  (experts, rev_of, skills)={ edge_index=[2, 135690] }
)

and the prepared model looks like this: -

Model(
  (loc_emb): Embedding(71, 64)
  (ex_emb): Embedding(13631, 64)
  (sk_emb): Embedding(69679, 64)
  (gnn): GraphModule(
    (conv1): ModuleDict(
      (location__of__experts): SAGEConv(64, 64, aggr=mean)
      (skills__of__experts): SAGEConv(64, 64, aggr=mean)
      (experts__rev_of__location): SAGEConv(64, 64, aggr=mean)
      (experts__rev_of__skills): SAGEConv(64, 64, aggr=mean)
    )
    (conv2): ModuleDict(
      (location__of__experts): SAGEConv(64, 64, aggr=mean)
      (skills__of__experts): SAGEConv(64, 64, aggr=mean)
      (experts__rev_of__location): SAGEConv(64, 64, aggr=mean)
      (experts__rev_of__skills): SAGEConv(64, 64, aggr=mean)
    )
  )
)

I want to perform link prediction only on 1 edge, that is, (location__of__experts). When I run the model.forward(), it returns an error with the following stack trace: -

Error:

Traceback (most recent call last):
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch\fx\graph_module.py", line 277, in __call__
    raise e
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch\fx\graph_module.py", line 267, in __call__
    return super(self.cls, obj).__call__(*args, **kwargs)  # type: ignore[misc]
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "<eval_with_key>.1", line 14, in forward
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\nn\conv\sage_conv.py", line 131, in forward
    out = self.propagate(edge_index, x=x, size=size)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\nn\conv\message_passing.py", line 467, in propagate
    out = self.aggregate(out, **aggr_kwargs)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\nn\conv\message_passing.py", line 591, in aggregate
    return self.aggr_module(inputs, index, ptr=ptr, dim_size=dim_size,
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\nn\aggr\base.py", line 131, in __call__
    return super().__call__(x, index, ptr, dim_size, dim, **kwargs)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\nn\aggr\basic.py", line 34, in forward
    return self.reduce(x, index, ptr, dim_size, dim, reduce='mean')
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\nn\aggr\base.py", line 170, in reduce
    return scatter(x, index, dim=dim, dim_size=dim_size, reduce=reduce)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\utils\scatter.py", line 82, in scatter
    return ScatterHelpers.scatter_mean(src, index, dim, out, dim_size)
  File "C:\Users\Arman\anaconda3\envs\pyg6\lib\site-packages\torch_geometric\utils\scatter.py", line 39, in scatter_mean
    out.scatter_add_(dim, index, src)
RuntimeError: index -128 is out of bounds for dimension 0 with size 13631

I tried debugging but wasn't able to resolve the error. Any idea where to look at to resolve this or what steps to follow to identify the mistake? Any help would be greatly appreciated.

Thanks.

Best,
Karan

[rusty1s]

Can you run data.validate() to see if your data is correctly formatted?

[karan96]

I checked my data using data.validate() and it returned that some of the data in had negative values. I verified my data to check the offending values and it worked. Thank you so much.