multi-dimensional edge_attr in add_self_loops()

[LanceKnight]

Hello,

I'm trying to inplement GINEconv. A simple thing now is to add self loop. I tested with a simple case but it didn't work. It looks like pyG does not support adding self loop with multi-dimential edge_attr?

from torch_geometric.utils import add_self_loops
import torch
edge_index = torch.tensor([[1, 0], [0, 1]])
edge_attr = torch.tensor([[1,1], [1,2]])
edge_index, edge_attr = add_self_loops(edge_index, edge_attr, 1)

Traceback (most recent call last):
File "temp.py", line 5, in
edge_index, edge_attr = add_self_loops(edge_index, edge_attr, 1)
File "/home/liuy69/bin/anaconda2/envs/graphormer/lib/python3.6/site-packages/torch_geometric/utils/loop.py", line 90, in add_self_loops
assert edge_weight.numel() == edge_index.size(1)
AssertionError

[LanceKnight]

nvm. I found I was using an old version of pyg