Issue with torch-sparse

[miriamcarnot]

Dear authors,
thank you for your work. I am trying to run the training and I installed all the dependencies as mentioned in the readme.
I am encountering the following error:

Traceback (most recent call last):
  File "/home/carnot/projects/UDA/saluda/train_single_back.py", line 5, in <module>
    from general_imports import *
  File "/home/carnot/projects/UDA/saluda/general_imports.py", line 8, in <module>
    import torch_geometric.transforms as T
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_geometric/__init__.py", line 4, in <module>
    import torch_geometric.data
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_geometric/data/__init__.py", line 1, in <module>
    from .data import Data
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_geometric/data/data.py", line 3, in <module>
    from torch_geometric.typing import OptTensor, NodeType, EdgeType
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_geometric/typing.py", line 4, in <module>
    from torch_sparse import SparseTensor
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_sparse/__init__.py", line 40, in <module>
    from .tensor import SparseTensor  # noqa
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_sparse/tensor.py", line 13, in <module>
    class SparseTensor(object):
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch/jit/_script.py", line 1302, in script
    _compile_and_register_class(obj, _rcb, qualified_name)
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch/jit/_recursive.py", line 44, in _compile_and_register_class
    script_class = torch._C._jit_script_class_compile(qualified_name, ast, defaults, rcb)
RuntimeError:
object has no attribute sparse_csc_tensor:
  File "/home/carnot/miniconda3/envs/torchsparse/lib/python3.10/site-packages/torch_sparse/tensor.py", line 585
            value = torch.ones(self.nnz(), dtype=dtype, device=self.device())

        return torch.sparse_csc_tensor(colptr, row, value, self.sizes())
               ~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE

torch_geometric requires me to install torch-sparse which is a different library than torchsparse. Do you use both? And if yes, which version of torch-sparse and torch_geometric are you using?

Thank you for your help and best regards :)

[BjoernMichele]

Dear @miriamcarnot ,

Thank you for your interest in our work.

Unfortunately, I have no access any more to my former dockerfile and can't give you the exact version numbers that I have used. I just recall that I used the versions matching to my cuda (11.3) and pytorch (1.11.0) versions.

I shortly tested the repo with my current setup (based on PyTorch 2.0.0, Cuda 11.7 and this Torchsparse commit: https://github.com/mit-han-lab/torchsparse.git@69c1034ddb285798619380537802ea0ff03aeba6). There were a few changes needed in some transformation functons but which have highly probably not caused your error.

With this new setup I could start a training without any errors.

In this new setup I use the following libraries:
torch-cluster 1.6.3+pt20cu117
torch-geometric 2.6.1
torch-scatter 2.1.2+pt20cu117
torch-sparse 0.6.18+pt20cu117

The error that you report appears familiar to me, but I can't recall the source or solution for it.

I hope this nevertheless can help you a bit.

With best regards,
Bjoern