自动求导，计算图示意图及pytorch实现

pytorch实现

x1 = torch.tensor(3.0, requires_grad=True)
y1 = torch.tensor(2.0, requires_grad=True)
a = x1 ** 2
b = 3 * a
c = b * y1
c.backward()
print(x1.grad)
print(y1.grad)
print(x1.grad == 6 * x1 * y1)
print(y1.grad == 3 * (x1 ** 2))
1
2
3
4
5
6
7
8
9
10

输出为：
tensor(36.)
tensor(27.)
tensor(True)
tensor(True)

默认情况下，pytorch会累加梯度，每次backward()前，需要进行梯度清零

x.grad.zero_()
1