inplace operation problem

[Oceanusity]

Hello, thank you for the detailed readme which helps us a lot to run the code. Now, I face a replacement operation problem when I run the code. It will be great if you can provide me with some insights on how to deal with it. Thank you.

python -m train.pattern sampling.mode random sampling.redundancy 5 sampling.random_rate 0.35 sampling.batch_factor 8

Here is the traceback when I set torch.autograd.set_detect_anomaly(True).

File "/GNNAK/core/model.py", line 37, in forward
    x = F.relu(x)
  File "/anaconda3/envs/exp/lib/python3.10/site-packages/torch/nn/functional.py", line 1457, in relu
    result = torch.relu(input)
 (Triggered internally at  /opt/conda/conda-bld/pytorch_1659484808560/work/torch/csrc/autograd/python_anomaly_mode.cpp:102.)
  Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
 
File "/GNNAK/train/pattern.py", line 85, in train
    loss.backward()
  File "/anaconda3/envs/exp/lib/python3.10/site-packages/torch/_tensor.py", line 396, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
  File "/anaconda3/envs/exp/lib/python3.10/site-packages/torch/autograd/__init__.py", line 173, in backward
    Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [54318, 144]], which is output 0 of ReluBackward0, is at version 1; expected version 0 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!

Here is my partial environment configuration:

ffmpeg                    4.3                  hf484d3e_0    pytorch
pyg                       2.1.0           py310_torch_1.12.0_cu113    pyg
pytorch                   1.12.1          py3.10_cuda11.3_cudnn8.3.2_0    pytorch
pytorch-cluster           1.6.0           py310_torch_1.12.0_cu113    pyg
pytorch-mutex             1.0                        cuda    pytorch
pytorch-scatter           2.0.9           py310_torch_1.12.0_cu113    pyg
pytorch-sparse            0.6.15          py310_torch_1.12.0_cu113    pyg
torchaudio                0.12.1              py310_cu113    pytorch
torchvision               0.13.1              py310_cu113    pytorch

[Oceanusity]

When I remove the line F.relu, this problem disappears.

[LingxiaoShawn]

Can you check whether you have updated to the newest package based on the instruction?
I can run the code successfully without error.
But it's true that the newest pytorch makes the in-place operation much harder,
you have to remove all in-place operation for correct back propagation.
I don't know why pytorch insists on this.

[zhouxuan1223]

I would like to ask if the environment you installed by running code directly installed the setup provided by the author. After I installed the latest setup provided by the author, I reported an error when running the environment

[LingxiaoShawn]

This should be cuda problem. Also the setup is using python 3.10.