Open gbg141 opened 2 weeks ago
You are correct that TUDataset.load
is not applicable due to additonal information being stored. That's why we use custom logic in TUDataset
to load the dataset. I personally think this is fine (since it is a rare case).
🛠Proposed Refactor
Update
InMemoryDataset
'sload()
method for it to be compatible with latest datasets (e.g.TUDatasets
) that store 4 elements in the processeddata.pd
object (data
,slices
,sizes
,data_cls
).Right now it is only compatible with 3 elements (
data
,slices
,data_cls
) or 2 (which is called backward compatibility, with onlydata
andslices
). When implementing a customInMemoryDataset
that might be of any size 2<=x<=4, the defaultload()
method is then obsolete with the newest options.Suggest a potential alternative/fix
An easy fix could be to allow 4 elements in the loading method of
InMemoryDataset
, something like: