SparseTensor.to_symmetric() inplace operation

[johannaSommer]

Hi,

thanks a lot for your work on torch_sparse! I have encountered the following error when computing gradients on the values of SparseTensor when it has previously been transformed with .to_symmetric(). I do not think this is intended as casting to a symmetric matrix should be a differentiable operation?

I was able to construct the following minimal example:

import torch
import torch_sparse

torch.manual_seed(0)
test_tensor = (torch.randn(5, 5) < 0.1).float()
test_tensor_sparse = torch_sparse.SparseTensor.from_dense(test_tensor)
test_tensor_sparse.requires_grad_()
test_tensor_sparse = test_tensor_sparse.to_symmetric()
loss = test_tensor_sparse.mean()
loss.backward()

Error:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_3483971/1228083090.py in <module>
      8 test_tensor_sparse = test_tensor_sparse.to_symmetric()
      9 loss = test_tensor_sparse.mean()
---> 10 loss.backward()

~/miniconda3/envs/molgen/lib/python3.7/site-packages/torch/_tensor.py in backward(self, gradient, retain_graph, create_graph, inputs)
    394                 create_graph=create_graph,
    395                 inputs=inputs)
--> 396         torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
    397 
    398     def register_hook(self, hook):

~/miniconda3/envs/molgen/lib/python3.7/site-packages/torch/autograd/__init__.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables, inputs)
    173     Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
    174         tensors, grad_tensors_, retain_graph, create_graph, inputs,
--> 175         allow_unreachable=True, accumulate_grad=True)  # Calls into the C++ engine to run the backward pass
    176 
    177 def grad(

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.LongTensor [30]] is at version 3; expected version 1 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

When setting torch.autograd.set_detect_anomaly(True):

...
 File "/nfs/homedirs/sommer/miniconda3/envs/molgen/lib/python3.7/site-packages/torch_sparse/tensor.py", line 373, in to_symmetric
    value = torch.cat([value, value])[perm]
 (Triggered internally at  /opt/conda/conda-bld/pytorch_1659484809535/work/torch/csrc/autograd/python_anomaly_mode.cpp:102.)
  allow_unreachable=True, accumulate_grad=True)  # Calls into the C++ engine to run the backward pass

This happens with the following specifications:

python: 3.7.13
torch: 1.12.1 
torch_sparse: 0.6.15

I hope this is helpful, let me know if you need more information or if I can help. Thanks!

[github-actions]

This issue had no activity for 6 months. It will be closed in 2 weeks unless there is some new activity. Is this issue already resolved?

[rusty1s]

@johannaSommer I am really sorry, I missed this issue. Taking a look now :(

[rusty1s]

Fixed via #327.