[博客精选]Neural Network using Python

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

A neural network trained with backpropagation is attempting to use input to predict output.

Inputs			Output
0	0	1	0
1	1	1	1
1	0	1	1
0	1	1	0

Consider trying to predict the output column given the three input columns. We could solve this problem by simply measuring statistics between the input values and the output values. If we did so, we would see that the leftmost input column is perfectly correlated with the output. Backpropagation, in its simplest form, measures statistics like this to make a model. Let's jump right in and use it to do this.

本帖隐藏的内容

8.pdf (6.82 MB)

2016-8-19 01:39:45 上传

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏2 回帖

关键词：network Neural python Using Work Network

1. import numpy as np
   
2. 
   
3. # sigmoid function
   
4. 
   
5. def nonlin(x,deriv=False):
   
6. 
   
7. if(deriv==True):
   
8. 
   
9. return x*(1-x)
   
10. 
    
11. return 1/(1+np.exp(-x))
    
12. 
    
13. # input dataset
    
14. 
    
15. X = np.array([  [0,0,1],
    
16. 
    
17. [0,1,1],
    
18. 
    
19. [1,0,1],
    
20. 
    
21. [1,1,1] ])
    
22. 
    
23. # output dataset           
    
24. 
    
25. y = np.array([[0,0,1,1]]).T
    
26. 
    
27. # seed random numbers to make calculation
    
28. 
    
29. # deterministic (just a good practice)
    
30. 
    
31. np.random.seed(1)
    
32. 
    
33. # initialize weights randomly with mean 0
    
34. 
    
35. syn0 = 2*np.random.random((3,1)) - 1
    
36. 
    
37. for iter in xrange(10000):
    
38. 
    
39. # forward propagation
    
40. 
    
41. l0 = X
    
42. 
    
43. l1 = nonlin(np.dot(l0,syn0))
    
44. 
    
45. # how much did we miss?
    
46. 
    
47. l1_error = y - l1
    
48. 
    
49. # multiply how much we missed by the
    
50. 
    
51. # slope of the sigmoid at the values in l1
    
52. 
    
53. l1_delta = l1_error * nonlin(l1,True)
    
54. 
    
55. # update weights
    
56. 
    
57. syn0 += np.dot(l0.T,l1_delta)
    
58. 
    
59. print "Output After Training:"
    
60. 
    
61. print l1

复制代码

1. import numpy as np
   
2. 
   
3. def nonlin(x,deriv=False):
   
4. 
   
5. if(deriv==True):
   
6. 
   
7. return x*(1-x)
   
8. 
   
9. return 1/(1+np.exp(-x))
   
10. 
    
11. X = np.array([[0,0,1],
    
12. [0,1,1],
    
13. [1,0,1],
    
14. [1,1,1]])
    
15. 
    
16. y = np.array([[0],
    
17. [1],
    
18. [1],
    
19. [0]])
    
20. 
    
21. np.random.seed(1)
    
22. 
    
23. # randomly initialize our weights with mean 0
    
24. 
    
25. syn0 = 2*np.random.random((3,4)) - 1
    
26. 
    
27. syn1 = 2*np.random.random((4,1)) - 1
    
28. 
    
29. for j in xrange(60000):
    
30. 
    
31. # Feed forward through layers 0, 1, and 2
    
32. 
    
33. l0 = X
    
34. 
    
35. l1 = nonlin(np.dot(l0,syn0))
    
36. 
    
37. l2 = nonlin(np.dot(l1,syn1))
    
38. 
    
39. # how much did we miss the target value?
    
40. 
    
41. l2_error = y - l2
    
42. 
    
43. if (j% 10000) == 0:
    
44. 
    
45. print "Error:" + str(np.mean(np.abs(l2_error)))
    
46. 
    
47. 
    
48. # in what direction is the target value?
    
49. 
    
50. # were we really sure? if so, don't change too much.
    
51. 
    
52. l2_delta = l2_error*nonlin(l2,deriv=True)
    
53. 
    
54. 
    
55. # how much did each l1 value contribute to the l2 error (according to the weights)?
    
56. 
    
57. l1_error = l2_delta.dot(syn1.T)
    
58. 
    
59. 
    
60. # in what direction is the target l1?
    
61. 
    
62. # were we really sure? if so, don't change too much.
    
63. 
    
64. l1_delta = l1_error * nonlin(l1,deriv=True)
    
65. 
    
66. syn1 += l1.T.dot(l2_delta)
    
67. 
    
68. syn0 += l0.T.dot(l1_delta)

复制代码

kankan