python实现简单神经网络算法

时间:2022-05-03 21:48:41

python实现简单神经网络算法,供大家参考,具体内容如下

python实现二层神经网络

包括输入层和输出层

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
import numpy as np
 
#sigmoid function
def nonlin(x, deriv = False):
  if(deriv == True):
    return x*(1-x)
  return 1/(1+np.exp(-x))
 
#input dataset
x = np.array([[0,0,1],
       [0,1,1],
       [1,0,1],
       [1,1,1]])
 
#output dataset
y = np.array([[0,0,1,1]]).T
 
np.random.seed(1)
 
#init weight value
syn0 = 2*np.random.random((3,1))-1
 
for iter in xrange(100000):
  l0 = x             #the first layer,and the input layer 
  l1 = nonlin(np.dot(l0,syn0))  #the second layer,and the output layer
 
 
  l1_error = y-l1
 
  l1_delta = l1_error*nonlin(l1,True)
 
  syn0 += np.dot(l0.T, l1_delta)
print "outout after Training:"
print l1
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
import numpy as np
 
#sigmoid function
def nonlin(x, deriv = False):
  if(deriv == True):
    return x*(1-x)
  return 1/(1+np.exp(-x))
 
#input dataset
x = np.array([[0,0,1],
       [0,1,1],
       [1,0,1],
       [1,1,1]])
 
#output dataset
y = np.array([[0,0,1,1]]).T
 
np.random.seed(1)
 
#init weight value
syn0 = 2*np.random.random((3,1))-1
 
for iter in xrange(100000):
  l0 = x             #the first layer,and the input layer 
  l1 = nonlin(np.dot(l0,syn0))  #the second layer,and the output layer
 
 
  l1_error = y-l1
 
  l1_delta = l1_error*nonlin(l1,True)
 
  syn0 += np.dot(l0.T, l1_delta)
print "outout after Training:"
print l1

这里,
l0:输入层

l1:输出层

syn0:初始权值

l1_error:误差

l1_delta:误差校正系数

func nonlin:sigmoid函数

python实现简单神经网络算法

可见迭代次数越多,预测结果越接近理想值,当时耗时也越长。

python实现三层神经网络

包括输入层、隐含层和输出层

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
import numpy as np
 
def nonlin(x, deriv = False):
  if(deriv == True):
    return x*(1-x)
  else:
    return 1/(1+np.exp(-x))
 
#input dataset
X = np.array([[0,0,1],
       [0,1,1],
       [1,0,1],
       [1,1,1]])
 
#output dataset
y = np.array([[0,1,1,0]]).T
 
syn0 = 2*np.random.random((3,4)) - 1 #the first-hidden layer weight value
syn1 = 2*np.random.random((4,1)) - 1 #the hidden-output layer weight value
 
for j in range(60000):
  l0 = X            #the first layer,and the input layer 
  l1 = nonlin(np.dot(l0,syn0)) #the second layer,and the hidden layer
  l2 = nonlin(np.dot(l1,syn1)) #the third layer,and the output layer
 
 
  l2_error = y-l2    #the hidden-output layer error
 
  if(j%10000) == 0:
    print "Error:"+str(np.mean(l2_error))
 
  l2_delta = l2_error*nonlin(l2,deriv = True)
 
  l1_error = l2_delta.dot(syn1.T)   #the first-hidden layer error
 
  l1_delta = l1_error*nonlin(l1,deriv = True)
 
  syn1 += l1.T.dot(l2_delta)
  syn0 += l0.T.dot(l1_delta)
print "outout after Training:"
print l2
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
import numpy as np
 
def nonlin(x, deriv = False):
  if(deriv == True):
    return x*(1-x)
  else:
    return 1/(1+np.exp(-x))
 
#input dataset
X = np.array([[0,0,1],
       [0,1,1],
       [1,0,1],
       [1,1,1]])
 
#output dataset
y = np.array([[0,1,1,0]]).T
 
syn0 = 2*np.random.random((3,4)) - 1 #the first-hidden layer weight value
syn1 = 2*np.random.random((4,1)) - 1 #the hidden-output layer weight value
 
for j in range(60000):
  l0 = X            #the first layer,and the input layer 
  l1 = nonlin(np.dot(l0,syn0)) #the second layer,and the hidden layer
  l2 = nonlin(np.dot(l1,syn1)) #the third layer,and the output layer
 
 
  l2_error = y-l2    #the hidden-output layer error
 
  if(j%10000) == 0:
    print "Error:"+str(np.mean(l2_error))
 
  l2_delta = l2_error*nonlin(l2,deriv = True)
 
  l1_error = l2_delta.dot(syn1.T)   #the first-hidden layer error
 
  l1_delta = l1_error*nonlin(l1,deriv = True)
 
  syn1 += l1.T.dot(l2_delta)
  syn0 += l0.T.dot(l1_delta)
print "outout after Training:"
print l2

以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持服务器之家。

原文链接:http://blog.csdn.net/stoneyyhit/article/details/52335468