Autograd.detect_anomaly() in CNN

autograd
1

Hi, I am using autograd.detect_anomaly() on my CNN and getting following output and I didn’t get it.

/usr/local/lib/python3.6/dist-packages/torch/autograd/init.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables) 97 Variable._execution_engine.run_backward( 98 tensors, grad_tensors, retain_graph, create_graph, —> 99 allow_unreachable=True) # allow_unreachable flag 100 101

RuntimeError: Function ‘PowBackward0’ returned nan values in its 0th output.

2

You get this output because one of the pow (power) operation returned nan values. Since you usually don’t want nan value, this is most likely a problem somewhere.