How to get the reported accuracy on Cora?

[hengruizhang98]

Hi, I am trying to reproduce the experiments on Cora dataset, but I cannot get the expected accuracy(> 86%) on Cora. Would you please give detailed instructions on how to get it?

[chaojiewang94]

I have the same problem with @hengruizhang98 . I didnot modify any code and the released code can only achieve <84% accuracy on Cora dataset.

[Zbiang]

I have the same problem with @hengruizhang98 . I didnot modify any code and the released code can only achieve <84% accuracy on Cora dataset.

Hi, How did you download the cora dataset, I tried various methods but failed to download.

[wiredosprout]

I have the same problem with @hengruizhang98 . I didnot modify any code and the released code can only achieve <84% accuracy on Cora dataset.

Hi, How did you download the cora dataset, I tried various methods but failed to download.

@Zbiang you can go with below 2 lines, at least I succeed with them. from dgl.data import citation_graph as citegrh ds = citegrh.load_cora()

[wzfhaha]

The split of cora dataset in dgl is different from the public split, which is pointed in https://github.com/kavehhassani/mvgrl/issues/2.

I have the same problem with @hengruizhang98 . I didnot modify any code and the released code can only achieve <84% accuracy on Cora dataset.

Hi, How did you download the cora dataset, I tried various methods but failed to download.

@Zbiang you can go with below 2 lines, at least I succeed with them. from dgl.data import citation_graph as citegrh ds = citegrh.load_cora()

[ltz0120]

I got the same problem with the CORA results. I tried a lot of hyper-parameters, but most of them give me <84% accuracy, and only a few can reach <85% with good luck. I wonder how to achieve 86.8% in the paper?