gradient computation

[hubinontheway]

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation ,may i ask why,thank you!

[mounchiliu]

Hi there,
The problem can be found by setting 'torch.autograd.set_detect_anomaly(True)' in your run_s-agcn.py, which can provide related traceback errors for this situation.
To solve this situation, I just replaced return self.relu(x) with return self.relu(x).clone().
I suppose this situation is caused because of the environment, i.e. different versions of pytorch or cuda.

[gan2360]

@mounchiliu .Your solution is useful，thank you for sharing!