Autograd: Automatic differential

Autograd

1、深度学习的算法本质上是通过反向传播求导数，Pytorch的Autograd模块实现了此功能；在Tensor上的所有操作，Autograd都能为他们自动提供微分，避免手动计算导数的复杂过程。2、autograd.Variable是Autograd中的核心类，它简单的封装了Tensor，并支持几乎所有Tensor操作；Tensor被封装为Variable之后，可以调用它的.backward()实现反向传播，自动计算所有的梯度。3、Variable主要包含三个属性：        data：保存Variable所包含的Tensor；        grad：保存data对应的梯度，grad也是个Variable，而不是Tensor，它和data的形状一样；        grad_fn：指向一个Function对象，这个Function用来反向传播计算输入的梯度。

Specific code parsing

2. #_Author_: Monkey
4. #!/usr/bin/env python
6. #-*-Coding:utf-8-*-
8. Import Torch as T
10. From Torch.autograd import Variable
14. x = Variable (T.ones (2,2), Requires_grad = True)
16. Print (x)
18. "'"tensor ([[1., 1.],
20. [1., 1.]], requires_grad=true) "
22. y = X.sum ()
24. Print (y)
26. "'tensor (4., grad_fn=<sumbackward0>) '
28. Print (Y.GRAD_FN) #指向一个Function对象 This function is used to reverse-propagate the gradient of the computed input
30. "'<sumbackward0 object at 0x000002d4240ab860> '
32. Y.backward ()
34. Print (X.grad)
36. "'"tensor ([[1., 1.],
38. [1., 1.]]
40. Y.backward ()
42. Print (X.grad)
44. "'"tensor ([[2., 2.],
46. [2., 2.]]
48. Y.backward ()
50. Print (X.grad)
52. "'"tensor ([[3., 3.],
54. [3., 3.]]
56. "'"grad in the process of reverse propagation (accumulated), which means running
58. Reverse propagation, the gradient will accumulate before the gradient, so the reverse propagation requires a gradient clear 0 "
60. Print (X.grad.data.zero_ ())
62. "'"tensor ([[0., 0.],
64. [0., 0.]]
68. Y.backward ()
70. Print (X.grad)
72. "'"tensor ([[1., 1.],
74. [1., 1.]]
78. m = Variable (T.ones (4,5))
80. n = T.cos (m)
82. Print (m)
84. Print (n)
86. "'"tensor ([[1., 1., 1., 1., 1.],
88. [1., 1., 1., 1., 1.],
90. [1., 1., 1., 1., 1.],
92. [1., 1., 1., 1., 1.]]
94. Tensor ([[0.5403, 0.5403, 0.5403, 0.5403, 0.5403],
96. [0.5403, 0.5403, 0.5403, 0.5403, 0.5403],
98. [0.5403, 0.5403, 0.5403, 0.5403, 0.5403],
100. [0.5403, 0.5403, 0.5403, 0.5403, 0.5403]] "
102. M_tensor_cos = T.cos (m.data)
104. Print (M_tensor_cos)
106. "'"Ensor ([0.5403, 0.5403, 0.5403, 0.5403, 0.5403],
108. [0.5403, 0.5403, 0.5403, 0.5403, 0.5403],
110. [0.5403, 0.5403, 0.5403, 0.5403, 0.5403],
112. [0.5403, 0.5403, 0.5403, 0.5403, 0.5403]] "

Autograd: Automatic differential