Issue loading data generated by new version of PyG

[davidireland3]

I have a data set of Data objects from the new version of PyTorch. I Pickled these on my workstation (which is using the latest version of PyG) and send them to my local machine. When I tried loading the data up using pickle on my local machine, which uses the previous version of PyG, I get the error ModuleNotFoundError: No module named 'torch_geometric.data.storage. I can't really update my PyG version on my local station as I have lots of these datasets generated using the previous version of PyG, so I am wondering whether there is a way around this so that I can load my new dataset in the old version of PyG?

Thanks!

[rusty1s]

You can save and load them via:

torch.save(data.to_dict(), path)
data = Data.from_dict(torch.load(path))

[davidireland3]

does this work if I have a list of data objects? should have clarified that in the original question, sorry.

e: assuming not as I don't believe a list has a .to_dict() method. I am assuming I can loop through my list and replace each Data object with .to_dict() and then pickle that corresponding list?

[rusty1s]

Yes, as long as you store dictionaries instead of individual data objects. You can do the conversion to dictionaries inside a list comprehension.

[davidireland3]

yep, that worked! thanks a lot for the quick reply! :-)