Why does Cora training set + validation Set + test set only use 1640 nodes instead of 2708？

[Chrisable]

In general,the number of the training nodes is 140,the validation Set is 500,the test nodes is 1000.why is not the sum 2708?When only using 1640 nodes,what is the benefits?

[Padarn]

Cora is implemented in a number of different datasets. Can you be more specific?

[rusty1s]

This is just a pre-defined split introduced in the GCN paper, where we only want to train on 20 examples per class. I don't think there is a real reason to not utilize all node labels during testing, but it is what it is :)