import torch

x = torch.tensor([2.], requires_grad=True)
w = torch.tensor([1.], requires_grad=True)

# 方式1
a = torch.add(w, x)
b = torch.add(w, 1)
y = torch.mul(a, b)

y.backward(retain_graph=True)
print(w.grad)  # tensor([5.])

w.grad.data.zero_()
y.backward(retain_graph=True)
print(w.grad)  # tensor([5.])

y.backward(retain_graph=True)
print(w.grad)  # tensor([10.])

y.backward(retain_graph=True)
print(w.grad)  # tensor([105.])

# 方式2
for _ in range(4):
    a = torch.add(w, x)
    b = torch.add(w, 1)
    y = torch.mul(a, b)

    w.grad.zero_()
    y.backward()
    print(w.grad)

"""
y.backward()
y.backward() 2次執行 backward()會報錯
RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed). 
Saved intermediate values of the graph are freed when you call .backward() or autograd.grad(). 
Specify retain_graph=True if you need to backward through the graph a second time or if you need to access saved tensors after calling backward.

"""