Pytorch_geometric: Dataset dimensions

[SaschaStenger]

❓ Questions & Help

Hello and sorry, that i have to ask again concerning dataset creation.

My current problem is, that i have my data object in the following form:
x.shape = [4289, 438] edge_index.shape = [4390, 2] edge_attr.shape = [4390]
x.shape = [3193, 438] edge_index.shape = [3804, 2] edge_attr.shape = [3804]

Now when the dataset comes to data, slices = self.collate(data_list)

I get a RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 4390 and 3804 in dimension 0 at C:\w\1\s\tmp_conda_3.7_055457\conda\conda-bld\pytorch_1565416617654\work\ate n\src\TH/generic/THTensor.cpp:689

But as far as it only tells me, that my tensors should match in dimension 1, which they do. They are both at 438. So why do i still get an error? (Sorry again if it's a basic question, but i just can't seam to figure it out)

Could this be, because i have more edges, then nods? This results, because some of my edges do have multiple attributes.

[xandox]

edge_index.shape must be [2, *]
in your case [2, 4390] and [2, 3804]

[SaschaStenger]

Thank you very much for the quick help

[xandox]

I have not finished yet :) Just have been looking for documentation.

You can transform your edge_index to correct format by calling edge_index.t().contiguous()

Example:

import torch
from torch_geometric.data import Data

edge_index = torch.tensor([[0, 1],
                           [1, 0],
                           [1, 2],
                           [2, 1]], dtype=torch.long)
x = torch.tensor([[-1], [0], [1]], dtype=torch.float)

data = Data(x=x, edge_index=edge_index.t().contiguous())
>>> Data(edge_index=[2, 4], x=[3, 1])

from here: https://pytorch-geometric.readthedocs.io/en/latest/notes/introduction.html?highlight=contiguous#data-handling-of-graphs

That is it. :)

[SaschaStenger]

Thanks. I never came back to the introduction by example and totally missed that.
Sorry for having to ask, but thank you for your patience in helping me.