PyTorch documentation has a note section for torch.optim.SGD optimizer that says:
torch.optim.SGD
The implementation of SGD with…
(Actually I vaguely remember encountering the same problem in MRO 3.3, and solved it using similar method. Unfortunately I didn’t write it down.)
Note: MRO means Microsoft R Open (previously Revolution R)
Full name of the competition: Planet: Understanding the Amazon from Space — Use satellite data to track the human footprint in the Amazon rain forest