The 'data' object was created by an older version of PyG

[hkim716]

I have a version error when loading MyDataset like this :

RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_1903/3483153563.py in <module>
----> 1 print(data[0])

~/miniconda3/envs/pyg_172/lib/python3.7/site-packages/torch_geometric/data/data.py in __repr__(self)
    395     def __repr__(self) -> str:
    396         cls = self.__class__.__name__
--> 397         has_dict = any([isinstance(v, Mapping) for v in self._store.values()])
    398 
    399         if not has_dict:

~/miniconda3/envs/pyg_172/lib/python3.7/site-packages/torch_geometric/data/data.py in __getattr__(self, key)
    356         if '_store' not in self.__dict__:
    357             raise RuntimeError(
--> 358                 "The 'data' object was created by an older version of PyG. "
    359                 "If this error occurred while loading an already existing "
    360                 "dataset, remove the 'processed/' directory in the dataset's "

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.

When I load data I think it successfully get the entire dataset, which has 17226 data_#.pt

data = XRF1LofiDataset('./processed')
print(len(data))
-> 17226

However, when I tried to see print(data[0]), it had a problem.
MyDataset looks like this, but this Dataset class was generated from the other version of PyG though.

class MyDataset(Dataset):
    def __init__(self, root, transform=None,
                 pre_transform=None,
                 pre_filter=None
                 ):
        super().__init__(root, transform, pre_transform, pre_filter)

    @property
    def raw_file_names(self):
        ff = [f for f in os.listdir(self.raw_dir)]
        return ff

    @property
    def processed_file_names(self):
        ff = [f for f in os.listdir(self.processed_dir)]
        return ff

    def len(self):
        return len(self.raw_file_names)

    def get(self, idx):
        data = torch.load(os.path.join(self.raw_dir, f'data_{idx}.pt'))
        return data

How could I fix this issue? My current torch_geometric version is 2.0.4.

[rusty1s]

Sadly, the data format needed to change slightly between PyG 1.* and PyG 2.*. You can either just re-process the data or do the following:

• Install PyG 1.7.4
• Load the data as usual and re-save them as dictionaries: torch.save(torch.load('data.pt').to_dict(), 'data_dict.pt')
• Install PyG 2.0.4
• Load the data and convert it to new data objects: Data.from_dict(torch.load('data_dict.pt'))

[Beteasy]

PyG 1.7.2 works for me