Error occurred when importing Data, DataLoader from torch_geometric.data

[xche564]

RuntimeError Traceback (most recent call last)
Cell In[1], line 6
4 import pickle as pkl
5 import torch
----> 6 from torch_geometric.data import Data, DataLoader
8 from utils.utils import build_spanning_tree_edge, find_higher_order_neighbors, add_self_loops

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_geometric/init.py:4
1 from types import ModuleType
2 from importlib import import_module
----> 4 import torch_geometric.data
5 import torch_geometric.loader
6 import torch_geometric.transforms

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_geometric/data/init.py:1
----> 1 from .data import Data
2 from .hetero_data import HeteroData
3 from .temporal import TemporalData

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_geometric/data/data.py:20
18 import torch
19 from torch import Tensor
---> 20 from torch_sparse import SparseTensor
22 from torch_geometric.data.feature_store import (
23 FeatureStore,
24 FeatureTensorType,
25 TensorAttr,
26 _field_status,
27 )
28 from torch_geometric.data.graph_store import (
29 EDGE_LAYOUT_TO_ATTR_NAME,
30 EdgeAttr,
(...)
34 edge_tensor_type_to_adj_type,
35 )

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_sparse/init.py:39
31 if t_major != major:
32 raise RuntimeError(
33 f'Detected that PyTorch and torch_sparse were compiled with '
34 f'different CUDA versions. PyTorch has CUDA version '
35 f'{t_major}.{t_minor} and torch_sparse has CUDA version '
36 f'{major}.{minor}. Please reinstall the torch_sparse that '
37 f'matches your PyTorch install.')
---> 39 from .storage import SparseStorage # noqa
40 from .tensor import SparseTensor # noqa
41 from .transpose import t # noqa

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_sparse/storage.py:21
16 assert layout == 'coo' or layout == 'csr' or layout == 'csc'
17 return layout
20 @torch.jit.script
---> 21 class SparseStorage(object):
22 _row: Optional[torch.Tensor]
23 _rowptr: Optional[torch.Tensor]

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch/jit/_script.py:924, in script(obj, optimize, _frames_up, _rcb)
922 if _rcb is None:
923 _rcb = _jit_internal.createResolutionCallbackFromFrame(_frames_up + 1)
--> 924 _compile_and_register_class(obj, _rcb, qualified_name)
925 return obj
926 else:
927 # this is a decorated fn, and we need to the underlying fn and its rcb

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch/jit/_script.py:64, in _compile_and_register_class(obj, rcb, qualified_name)
62 ast = get_jit_class_def(obj, obj.name)
63 defaults = torch.jit.frontend.get_default_args_for_class(obj)
---> 64 torch._C._jit_script_class_compile(qualified_name, ast, defaults, rcb)
65 torch.jit._state._add_script_class(obj, qualified_name)

RuntimeError:
Arguments for call are not valid.
The following variants are available:

aten::div.Tensor(Tensor self, Tensor other) -> (Tensor):
Expected a value of type 'Tensor' for argument 'other' but instead found type 'int'.

aten::div.Scalar(Tensor self, Scalar other) -> (Tensor):
Keyword argument rounding_mode unknown.

aten::div.out(Tensor self, Tensor other, *, Tensor(a!) out) -> (Tensor(a!)):
Expected a value of type 'Tensor' for argument 'other' but instead found type 'int'.

aten::div.int(int a, int b) -> (float):
Keyword argument rounding_mode unknown.

aten::div.float(float a, float b) -> (float):
Expected a value of type 'float' for argument 'b' but instead found type 'int'.

aten::div(Scalar a, Scalar b) -> (float):
Keyword argument rounding_mode unknown.

div(float a, Tensor b) -> (Tensor):
Expected a value of type 'Tensor' for argument 'b' but instead found type 'int'.

div(int a, Tensor b) -> (Tensor):
Expected a value of type 'Tensor' for argument 'b' but instead found type 'int'.

The original call is:
File "/home/xuchen/.conda/envs/xcml/lib/python3.9/site-packages/torch_sparse/storage.py", line 316
idx = self.sparse_size(1) * self.row() + self.col()

    row = torch.div(idx, num_cols, rounding_mode='floor')
          ~~~~~~~~~ <--- HERE
    col = idx % num_cols
    assert row.dtype == torch.long and col.dtype == torch.long

I have had all the torch_geometric related packages installed.

Name Version Build Channel

_pytorch_select 0.1 cpu_0
pytorch 1.7.1 py3.9_cuda11.0.221_cudnn8.0.5_0 pytorch
torch 1.13.0 pypi_0 pypi
torch-cluster 1.6.0 pypi_0 pypi
torch-geometric 2.2.0 pypi_0 pypi
torch-scatter 2.0.7 pypi_0 pypi
torch-sparse 0.6.16 pypi_0 pypi
torch-spline-conv 1.2.1 pypi_0 pypi
torchaudio 0.7.2 py39 pytorch
torchvision 0.14.0 pypi_0 pypi

[rusty1s]

If you are running with PyTorch 1.7.1, can you try to install torch-sparse==0.6.9. Later versions are not necessarily compatible with older PyTorch versions :(

[xche564]

Thanks for your help! This does solve the original issue. But another one popped up.

ImportError Traceback (most recent call last)
Cell In[1], line 6
4 import pickle as pkl
5 import torch
----> 6 from torch_geometric.data import Data, DataLoader
8 from utils.utils import build_spanning_tree_edge, find_higher_order_neighbors, add_self_loops

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_geometric/init.py:9
7 import torch_geometric.utils
8 import torch_geometric.explain
----> 9 import torch_geometric.profile
11 from .seed import seed_everything
12 from .home import get_home_dir, set_home_dir

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_geometric/profile/init.py:1
----> 1 from .profile import profileit, timeit, get_stats_summary
2 from .profile import trace_handler, rename_profile_file, torch_profile
3 from .utils import count_parameters

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch_geometric/profile/profile.py:8
5 from typing import Any, List, NamedTuple, Tuple
7 import torch
----> 8 from torch.profiler import ProfilerActivity, profile
10 from torch_geometric.profile.utils import (
11 byte_to_megabyte,
12 get_gpu_memory_from_nvidia_smi,
13 )
16 class Stats(NamedTuple):

File ~/.conda/envs/xcml/lib/python3.9/site-packages/torch/profiler/init.py:10
1 r"""
2 PyTorch Profiler is a tool that allows the collection of performance metrics during training and inference.
3 Profiler's context manager API can be used to better understand what model operators are the most expensive,
(...)
8
9 """
---> 10 from torch._C._autograd import _supported_activities, DeviceType, kineto_available
11 from torch._C._profiler import _ExperimentalConfig, ProfilerActivity, RecordScope
12 from torch.autograd.profiler import record_function

ImportError: cannot import name '_supported_activities' from 'torch._C._autograd' (unknown location)

Is this still because of the mismatch of the versions of packages?

[rusty1s]

Yeah, looks like it. One way to resolve that would be to simply catch the exception in import torch_geometric - otherwise, you may need to downgrade PyG as well :(