欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

Pytorch 自动微分

程序员文章站 2022-07-12 23:03:07
...
  • Tensor.requires_grad = True 记录对Tensor的所有操作,后序.backward() 自动计算所有梯度到 .grad 属性
import torch
x = torch.ones(2,2, requires_grad=True) # 默认是False
print(x)

tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
  • 停止记录调用.detach()
x.detach_()
print(x.requires_grad) # False
  • .grad_fn 保存了创建张量的 Function 的引用
y = x + 2
print(y)
print(y.grad_fn)

tensor([[3., 3.],
        [3., 3.]], grad_fn=<AddBackward0>)
<AddBackward0 object at 0x0000015716529D68>
z = y*y*3
out = z.mean()
print(z, out)

tensor([[27., 27.],
        [27., 27.]], grad_fn=<MulBackward0>) 

tensor(27., grad_fn=<MeanBackward0>)
# requires_grad 默认为 False
a = torch.randn(2, 2)
a = ((a*3)/(a-1))
print(a.requires_grad)  # False
b = (a*a).sum()
print(b.grad_fn)  # None

a.requires_grad_(True)  # 设置为 True
print(a.requires_grad)  # True
b = (a*a).sum()
print(b.grad_fn)
# <SumBackward0 object at 0x0000015717DC69E8>
  • backward() 后向传播
z = y*y*3
y = x+2
计算 d(out)/dx

o u t = 1 4 ( ∑ 3 ( x i + 2 ) 2 ) → d o u t d x i = 3 2 ( x i + 2 ) out = \frac{1}{4}(\sum3(x_i+2)^2) \rightarrow \frac{d_{out}}{dx_i} = \frac{3}{2}(x_i+2) out=41(3(xi+2)2)dxidout=23(xi+2)
x i = 1 , d o u t / d x i = 4.5 x_i = 1, d_{out}/dx_i = 4.5 xi=1,dout/dxi=4.5

out.backward()
print(y.grad) # None, 为什么?是 None
print(x.grad)
tensor([[4.5000, 4.5000],
        [4.5000, 4.5000]])

J = ( ∂ y 1 ∂ x 1 ⋯ ∂ y m ∂ x 1 ⋮ ⋱ ⋮ ∂ y 1 ∂ x n ⋯ ∂ y m ∂ x n ) J=\left(\begin{array}{ccc}\frac{\partial y_{1}}{\partial x_{1}} & \cdots & \frac{\partial y_{m}}{\partial x_{1}} \\ \vdots & \ddots & \vdots \\ \frac{\partial y_{1}}{\partial x_{n}} & \cdots & \frac{\partial y_{m}}{\partial x_{n}}\end{array}\right) J=x1y1xny1x1ymxnym

  • 当又使用了一个函数 l = g ( y ) l = g(y) l=g(y),v 是 l l l y y y 的导数,链式求导相乘,得到 l l l x x x 的导数
    J ⋅ v = ( ∂ y 1 ∂ x 1 ⋯ ∂ y m ∂ x 1 ⋮ ⋱ ⋮ ∂ y 1 ∂ x n ⋯ ∂ y m ∂ x n ) ( ∂ l ∂ y 1 ⋮ ∂ l ∂ y m ) = ( ∂ l ∂ x 1 ⋮ ∂ l ∂ x n ) J \cdot v=\left(\begin{array}{ccc}\frac{\partial y_{1}}{\partial x_{1}} & \cdots & \frac{\partial y_{m}}{\partial x_{1}} \\ \vdots & \ddots & \vdots \\ \frac{\partial y_{1}}{\partial x_{n}} & \cdots & \frac{\partial y_{m}}{\partial x_{n}}\end{array}\right)\left(\begin{array}{c}\frac{\partial l}{\partial y_{1}} \\ \vdots \\ \frac{\partial l}{\partial y_{m}}\end{array}\right)=\left(\begin{array}{c}\frac{\partial l}{\partial x_{1}} \\ \vdots \\ \frac{\partial l}{\partial x_{n}}\end{array}\right) Jv=x1y1xny1x1ymxnymy1lyml=x1lxnl

上面代码改为:

v = torch.tensor(2, dtype=torch.float)
out.backward(v)
print(x.grad)

# 梯度乘以了 2
tensor([[9., 9.],
        [9., 9.]])
  • 评估阶段可以使用 with torch.no_grad(): 不需要梯度计算和更新
print(x.requires_grad) # True
print((x ** 2).requires_grad) # True

# 取消梯度记录
with torch.no_grad():
    print((x ** 2).requires_grad) # False
相关标签: Pytorch