Cora dataset node order shuffling

[cxw-droid]

Hi,

I want to shuffle the node order of cora data (edge_index, node label should be changed accordingly). Would you like to recommend a way to do it? It is easy to do node label shuffling accordingly, but changing edge_index seems more difficult. Thanks.

[EdisonLeeeee]

perm = torch.randperm(data.num_nodes)
data.x = data.x[perm]
data.y = data.y[perm]
data.edge_index = perm[data.edge_index]

[cxw-droid]

Thank you. It seems the new graph is different from the original one. The new edge_index seems not correct. E.g., if originally node 0, 1 is connected, then it becomes a new edge (perm[0], perm[1]). After permutation, we need to know the new number of original node 0, node 1 and then connect them.

[EdisonLeeeee]

The graph is still the same as the associated features and labels for nodes are the same. perm[0], perm[1] is just the new number of original 0 and 1 after permutation.

[cxw-droid]

The graph is still the same as the associated features and labels for nodes are the same. perm[0], perm[1] is just the new number of original 0 and 1 after permutation.

The original node numbers perm[0], perm[1] are changed to the new node numbers 0, 1, as shown in data.x = data.x[perm], but it seems we cannot say the original 0, 1 are changed to perm[0], perm[1]

[EdisonLeeeee]

Sorry, you are right. That should work now

perm = torch.randperm(data.num_nodes)
data.x[perm] = data.x.clone()
data.y[perm] = data.y.clone()
data.edge_index = perm[data.edge_index]

[cxw-droid]

Thank you. Another implementation:

perm = torch.randperm(data.num_nodes)
x = data.x[perm]
y = data.y[perm]
new_idx = torch.zeros_like(perm)
new_idx[perm] = torch.arange(data.num_nodes)
edge_index = new_idx[data.edge_index]