Questions about Cora dataset

[scottshufe]

1. First, when I use dataset = Planetoid(path, 'Cora')[0] to load Cora dataset, and

from torch_geometric.utils import to_networkx

G = to_networkx(data)
G = nx.to_undirected(G)
G = nx.Graph(G)
nx.info(G)

I got 'Graph with 2708 nodes and 5278 edges'. 5278 = 10556 / 2, there is no problem.

2. Second, when I use the RandomLinkSplit transformation T.RandomLinkSplit(num_val=0.05, num_test=0.1, is_undirected=True, add_negative_train_samples=False) to load Cora dataset = Planetoid(path, 'Cora', transform)[0] and I get the train_data (8976 edges), val_data, test_data. Then,

from torch_geometric.utils import to_networkx

G = to_networkx(train_data)
G = nx.to_undirected(G)
G = nx.Graph(G)
nx.info(G)

I got 'Graph with 2708 nodes and 3513 edges'

Then, my questions is why the edge num of the obtained graph (3513) does not equal to the half of train_data edge num (8976/2=4488)

[EdisonLeeeee]

Which PyG version are you using? This bug has been resolved for PyG>=2.0.4

[scottshufe]

My PyG version is 2.0.2. I'll check it by newer PyG version that >= 2.0.4. Thanks.

[scottshufe]

I have checked this bug for PyG == 2.1.0 and it has been resolved. Thanks for your help.