pytorch 反向传播backward

设置变量x1, x2, x3
设置函数
z = x1 * x2 + x3
并且对x1进行求导，所以求出的结果因该为x2

 

requires_grad=True //设置变量是可以求导的
z.backward()//是z进行反向传播
retain_graph = True//可以进行多次的求导

x1 = torch.randn(1, requires_grad=True)
print("x1", x1)

x2 = torch.randn(1, requires_grad=True)
print("x2", x2)

x3 = torch.randn(1, requires_grad=True)
print("x3", x3)

y = x1*x2
z = y+x3
print(z.requires_grad)
print('z', z)

# 梯度如果不清空，每次循环就会累加
z.backward()
# print(y.grad)
print(x1.grad)

x1 = torch.randn(1, requires_grad=True)
print("x1", x1)

x2 = torch.randn(1, requires_grad=True)
print("x2", x2)

x3 = torch.randn(1, requires_grad=True)
p