# Calculating gradients
xt = tensor([3.,4.,10.]).requires_grad_()
xt
---------------------------------
tensor([ 3., 4., 10.], requires_grad=True)
1
2
3
4
5
2
3
4
5
The requires_grad_() method tells PyTorch to tag that variable at that value so that we can calculate its gradient at a later time.
def f(x): return x**2
yt = f(xt)
yt
_________________________________
tensor(9., grad_fn=<PowBackward0>)
1
2
3
4
5
6
2
3
4
5
6
# Calculate the gradient
yt.backward()
1
# View the gradient by accessing the grad property
xt.grad
1