benchmarking-gnns/data/TUs at master · bdy9527/benchmarking-gnns · GitHub

Name		Name	Last commit message	Last commit date
parent directory ..
COLLAB_test.index		COLLAB_test.index
COLLAB_train.index		COLLAB_train.index
COLLAB_val.index		COLLAB_val.index
DD_test.index		DD_test.index
DD_train.index		DD_train.index
DD_val.index		DD_val.index
ENZYMES_test.index		ENZYMES_test.index
ENZYMES_train.index		ENZYMES_train.index
ENZYMES_val.index		ENZYMES_val.index
PROTEINS_full_test.index		PROTEINS_full_test.index
PROTEINS_full_train.index		PROTEINS_full_train.index
PROTEINS_full_val.index		PROTEINS_full_val.index
README.md		README.md

README.md

Details:

Split total number of graphs into 3 (train, val and test) in 80:10:10
Stratified split proportionate to original distribution of data with respect to classes
Using sklearn to perform the split and then save the indexes
Preparing 10 such combinations of indexes split to be used in Graph NNs
As with KFold, each of the 10 fold have unique test set.