train test split question

[yyysyyy]

I found that in the train_test_split function in gwnn.py, you divide 80% of the nodes in the cora data set into training sets for training. Is this in line with common sense? Or I misunderstood or your code is wrong？