RuntimeError: The 'data' object was created by an older version of PyG. If this error occurred while loading an already existing dataset, remove the 'processed/' directory in the dataset's root folder and try again.

[zwangeo]

I was utilizing the latest version of python-geometric 2.X to achieve some latest implementations. When I have to handle the dataset processed by the older version of PyG (1.X), this error occurs. I'd like to inquire, is there a way that I can handle the data without degrading the PyG version to 1.X (since some lastest implementations in 2.X is really needed)? The following is the full traceback. Many thanks!!

Traceback (most recent call last):
File "/export/data/zhili/projects/4th/ours/pretrain-gnns/bio/finetune.py", line 253, in
main()
File "/export/data/zhili/projects/4th/ours/pretrain-gnns/bio/finetune.py", line 178, in main
trainval_dataset, test_dataset = species_split(dataset)
File "/export/data/zhili/projects/4th/ours/pretrain-gnns/bio/splitters.py", line 66, in species_split
train_valid_dataset = dataset[train_valid_byte_tensor.tolist()]
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/dataset.py", line 263, in getitem
return self.index_select(idx)
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/dataset.py", line 298, in index_select
dataset = copy.copy(self)
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/copy.py", line 92, in copy
rv = reductor(4)
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/in_memory_dataset.py", line 168, in getattr
if isinstance(data, Data) and key in data:
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/data.py", line 145, in contains
return key in self.keys
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/data.py", line 134, in keys
for store in self.stores:
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/data.py", line 499, in stores
return [self._store]
File "/export/data/zhili/anaconda3/envs/fourth/lib/python3.9/site-packages/torch_geometric/data/data.py", line 436, in getattr
raise RuntimeError(
RuntimeError: The 'data' object was created by an older version of PyG. If this error occurred while loading an already existing dataset, remove the 'processed/' directory in the dataset's root folder and try again.

[rusty1s]

Sadly, you currently have to downgrade PyG, really sorry. I will look into ways to make this more performant across different releases. With downgrading, you can then do

In PyG 1.*
data = torch.load('data.pt')
torch.save(data.to_dict(), 'data_dict.pt')

# In PyG 2.*
data_dict = torch.load('data_dict.pt')
data = Data.from_dict(data_dict)
torch.save(data, 'data.pt')

[zwangeo]

Thank you very much for the instant reply!

[jasperhyp]

Thank you, this is very helpful for me too. One note is that when installing the old torch_geometric, you might still want to install the newer torch_scatter or else there would likely be an OSError like OSError: libpython3.7m.so.1.0: cannot open shared object file: No such file or directory.

[rishabh-ranjan]

This worked for me without having to downgrade and then upgrade pyg:

# in pyg 2.*

def bump(g):
    return Data.from_dict(g.__dict__)

old_data = torch.load("data.pt")
new_data = bump(old_data)

@rusty1s are there any caveats to this?

[rusty1s]

This seems to be fine :)