Closed Jillian555 closed 1 year ago
Thanks for your interest!
We didn't quite understand your question. But we could try to make some clarification to see whether it resolves your question.
The cls_balance here indicates whether we want to weigh all classes equally (cls_balance=True) or not. If we set it as True, we weigh all classes equally, and the accuracy is averaged over all classes regardless of the size of each class. In this case, the mask parameter is not used in the accuracy function. Instead, the ids_per_class is used to provide the ids of the nodes in each class.
When we set is as False, then the accuracy is averaged over all nodes, i.e. larger classes would dominate the result. In this case, we do not need the ids_per_cls. Instead, the mask parameter is used to indicate which nodes are testing nodes. You are correct on this, After the mask parameter is set, the sample for testing is selected. That is what we want to do.
Thanks for your answering!
On lines 95-98, the cls_balance parameter should be set instead of the mask parameter. After the mask parameter is set, the sample for testing is selected.