Hi @steelman97 ,
I think writing a new dataset for node tasks is better. Please refer to the class MyGraphPropPredDataset if you want to do so.
There are some tips for trying the model on ogbn-proteins:
Implement a proper graph sampling strategy as mentioned in our paper.
Some parts of the code need to rewrite to fit for larger graph. (E.g., the floyd-warshall algorithm is n^3 regarding the number of nodes, you may need to choose an alternative metric for efficient spatial encoding.)
What do I have to modify the code in order to try the model on ogbn-proteins? (Wrapper, collator..)