[人工智能]Pytorch的自动微分

感谢：Pytorch深度学习实战
出版社：清华大学出版社

import numpy as np
import torch

Tensor的requires_grad属性设置为True时，Pytorch的torch.autograd会自动地追踪它的计算轨迹，当需要计算微分的时候，只需要对最终计算的tensor调用backward方法，中间所有的计算节点的微分就会被保存在grad属性。

a = torch.arange(9).view(3,3)
a.requires_grad

False

x = torch.rand(3,3,requires_grad  = True)
print('微分后的X\n：',x)

微分后的X
： tensor([[0.4568, 0.3014, 0.1629],
        [0.9983, 0.9308, 0.4588],
        [0.7727, 0.1441, 0.9768]], requires_grad=True)

w = torch.ones(3,3,requires_grad = True)
y = torch.sum(torch.mm(x,w))

print(y)

tensor(15.6081, grad_fn=<SumBackward0>)

# 反向更新x\w
print(y.backward)
print(y.grad)
print(x.grad)
print(w.grad)

<bound method Tensor.backward of tensor(15.6081, grad_fn=<SumBackward0>)>
None
tensor([[3., 3., 3.],
        [3., 3., 3.],
        [3., 3., 3.]])
tensor([[2.2278, 2.2278, 2.2278],
        [1.3763, 1.3763, 1.3763],
        [1.5986, 1.5986, 1.5986]])


D:\install_file\Anaconda3\envs\5tsp\lib\site-packages\torch\_tensor.py:1013: UserWarning: The .grad attribute of a Tensor that is not a leaf Tensor is being accessed. Its .grad attribute won't be populated during autograd.backward(). If you indeed want the .grad field to be populated for a non-leaf Tensor, use .retain_grad() on the non-leaf Tensor. If you access the non-leaf Tensor by mistake, make sure you access the leaf Tensor instead. See github.com/pytorch/pytorch/pull/30531 for more informations. (Triggered internally at  aten\src\ATen/core/TensorBody.h:417.)
  return self._grad

# Tensor.detach会将Tensor从计算图中剥离出去，不在计算他的微分，
x = torch.rand(3,3,requires_grad = True)
w = torch.rand(3,3,requires_grad = True)
print('x:',x)
print('w:',w)
yy = torch.mm(w,x)
detach_yy = yy.detach()
y = torch.mean(yy)
y.backward()
print('-'*60)
print(yy.grad)
print(detach_yy)
print(x.grad)
print(w.grad)

# with torch.no_grad():包括的代码段不会计算微分
print('-'*60)
y = torch.sum(torch.mm(w,x))
print(y.requires_grad)

with torch.no_grad():
    y = torch.sum(torch.mm(w,x))
    print(y.requires_grad)

x: tensor([[0.2699, 0.2150, 0.4124],
        [0.5506, 0.1069, 0.6992],
        [0.1359, 0.9054, 0.6022]], requires_grad=True)
w: tensor([[0.3951, 0.9469, 0.1590],
        [0.0279, 0.3368, 0.9199],
        [0.0966, 0.1213, 0.7982]], requires_grad=True)
------------------------------------------------------------
None
tensor([[0.6496, 0.3302, 0.9208],
        [0.3180, 0.8749, 0.8009],
        [0.2013, 0.7564, 0.6053]])
tensor([[0.0577, 0.0577, 0.0577],
        [0.1561, 0.1561, 0.1561],
        [0.2086, 0.2086, 0.2086]])
tensor([[0.0997, 0.1507, 0.1826],
        [0.0997, 0.1507, 0.1826],
        [0.0997, 0.1507, 0.1826]])
------------------------------------------------------------
True
False