Why do RandomLinkSplit and RandomNodeSplit have different types of outputs?

[meredith-martz]

Is there a reason that RandomNodeSplit adds train_mask, val_mask and test_mask attributes to the Data object whereas RandomLinkSplit returns three different subgraphs? It is confusing as the end user that the calling behavior is different for the two functions. I would expect it to be consistent across the two methods.

[rusty1s]

I agree, it‘s sadly not consistent. In general, it is preferred to only have one data object with mask attributes (e.g., to share data more easily across different splits). However, this is not possible in random link splits as the graph actually changes across splits. There is an option to only have one data attribute in random link splits as well though (see split_labels argument).

[meredith-martz]

Fair enough!