Potential error in training loop

[sholtoarmstrong-iot]

In line train.py:100, The "loss.backward()" inside the sub_iter loop. However in line train.py:68 the "optimizer.zero_grad()" is applied outside the loop.

I am not sure if this is intended as to my knowledge the zero grad should be done between each backwards operation? This means the zero_grad should either be inside the sub_iter loop or the loss.backwards() should be replaced with a batch_loss.backwards() outside of the loop.