在线顺序极限学习机算法

   

实验：

[python] view plain copy

print ?

1. # coding:UTF-8  
2. #################  
3. # OS_ELM  
4. # author : zhiyong_will  
5. # date : 2015.3.22  
6. #################  
7. from __future__ import division  
8. from datetime import datetime  
9. from csv import DictReader  
10. from math import exp  
11. import random  
12. from numpy import *  
13. import string  
14.   
15. #####设置相关参数########  
16. trainData = ”C:\\Users\\dell\\Desktop\\OS-ELM\\segment_train.csv”  
17. testData = ”C:\Users\dell\Desktop\OS-ELM\\segment_test.csv”  
18. #隐含神经元的个数  
19. nHiddenNeurons = 180  
20. #输入层的神经元个数  
21. nInputNeurons = 19  
22. #初始训练集的大小  
23. NO = 280  
24.   
25. #函数  
26. def sig(tData, Iw, bias, num):  
27.     ””’ 
28.     tData:样本矩阵：样本数*特征数 
29.     Iw:输入层到第一个隐含层的权重：隐含层神经元数*特整数 
30.     bias:偏置1*隐含神经元个数 
31.     ”’  
32.     v = tData * Iw.T   #样本数*隐含神经元个数  
33.     bias_1 = ones((num, 1)) * bias  
34.     v = v + bias_1  
35.     H = 1./(1+exp(-v))  
36.     return H  
37.       
38.   
39. ##导入数据集  
40. firstTrainData = []  
41. firstTrainLable = []  
42.   
43. # 处理训练样本  
44. for t, row in enumerate(DictReader(open(trainData))):  
45.     Id = row[’Id’]  
46.     del row[‘Id’]  
47.     del row[‘I0’]  
48.       
49.     data = []  
50.     if int(Id) < NO:  
51.         # 处理是否被点击  
52.         if row[‘Label’] == ‘1.00000000’:  
53.             y = 1  
54.         elif row[‘Label’] == ‘2.00000000’:  
55.             y = 2  
56.         elif row[‘Label’] == ‘3.00000000’:  
57.             y = 3  
58.         elif row[‘Label’] == ‘4.00000000’:  
59.             y = 4  
60.         elif row[‘Label’] == ‘5.00000000’:  
61.             y = 5  
62.         elif row[‘Label’] == ‘6.00000000’:  
63.             y = 6  
64.         else:  
65.             y = 7          
66.         del row[‘Label’]  
67.         firstTrainLable.append(y)  
68.         # 处理特征  
69.         for key in row:  
70.             value = string.atof(row[key])  
71.             #index = int(value + key[1:], 16) % D  
72.             data.append(value)  
73.           
74.         firstTrainData.append(data)  
75.         continue  
76.     elif int(Id) == NO:#开始训练  
77.         p0 = mat(firstTrainData)  
78.         T0 = zeros((NO, 7))  
79.         #处理样本标签  
80.         for i in xrange(0, NO):   
81.             a = firstTrainLable[i]  
82.             T0[i][a-1] = 1  
83.           
84.         T0 = T0 * 2 - 1  
85.         Iw = mat(random.rand(nHiddenNeurons, nInputNeurons) * 2 - 1)#随机生成区间-1,1之间的随机矩阵  
86.         bias = mat(random.rand(1, nHiddenNeurons))  
87.         H0 = sig(p0, Iw, bias, NO)#样本数*隐含神经元个数  
88.         M = (H0.T * H0).I  
89.         beta = M * H0.T * T0  
90.     else:#训练剩余的样本,每次训练一条样本  
91.         # 处理label  
92.         if row[‘Label’] == ‘1.00000000’:  
93.             y = 1  
94.         elif row[‘Label’] == ‘2.00000000’:  
95.             y = 2  
96.         elif row[‘Label’] == ‘3.00000000’:  
97.             y = 3  
98.         elif row[‘Label’] == ‘4.00000000’:  
99.             y = 4  
100.         elif row[‘Label’] == ‘5.00000000’:  
101.             y = 5  
102.         elif row[‘Label’] == ‘6.00000000’:  
103.             y = 6  
104.         else:  
105.             y = 7          
106.         del row[‘Label’]  
107.         Tn = zeros((1, 7))  
108.         #处理样本标签  
109.         b = y  
110.         Tn[0][b-1] = 1  
111.         Tn = Tn * 2 - 1  
112.         # 处理特征  
113.         data = []  
114.         for key in row:  
115.             value = string.atof(row[key])  
116.             data.append(value)  
117.         pn = mat(data)  
118.         H = sig(pn, Iw, bias, 1)  
119.         M = M - M * H.T * (eye(1,1) + H * M * H.T).I * H * M  
120.         beta = beta + M * H.T * (Tn - H * beta)  
121.   
122. # 计算训练误差  
123. correct = 0  
124. sum = 0  
125. for t, row in enumerate(DictReader(open(trainData))):  
126.     del row[‘Id’]  
127.     del row[‘I0’]  
128.       
129.     # 处理是否被点击  
130.     if row[‘Label’] == ‘1.00000000’:  
131.         y = 1  
132.     elif row[‘Label’] == ‘2.00000000’:  
133.         y = 2  
134.     elif row[‘Label’] == ‘3.00000000’:  
135.         y = 3  
136.     elif row[‘Label’] == ‘4.00000000’:  
137.         y = 4  
138.     elif row[‘Label’] == ‘5.00000000’:  
139.         y = 5  
140.     elif row[‘Label’] == ‘6.00000000’:  
141.         y = 6  
142.     else:  
143.         y = 7          
144.     del row[‘Label’]  
145.       
146.     # 处理特征  
147.     data = []  
148.     for key in row:  
149.         value = string.atof(row[key])  
150.         data.append(value)  
151.       
152.     p = mat(data)  
153.     HTrain = sig(p, Iw, bias, 1)  
154.     Y = HTrain * beta  
155.       
156.     # 判断  
157.     if argmax(Y) + 1 == y:  
158.         correct += 1  
159.     sum += 1  
160. print(“训练准确性为：%f” % (correct/sum))  
161.   
162. # 计算测试误差  
163. correctTest = 0  
164. sumTest = 0  
165. for t, row in enumerate(DictReader(open(testData))):  
166.     del row[‘Id’]  
167.     del row[‘I0’]  
168.       
169.     # 处理是否被点击  
170.     if row[‘Label’] == ‘1.00000000’:  
171.         y = 1  
172.     elif row[‘Label’] == ‘2.00000000’:  
173.         y = 2  
174.     elif row[‘Label’] == ‘3.00000000’:  
175.         y = 3  
176.     elif row[‘Label’] == ‘4.00000000’:  
177.         y = 4  
178.     elif row[‘Label’] == ‘5.00000000’:  
179.         y = 5  
180.     elif row[‘Label’] == ‘6.00000000’:  
181.         y = 6  
182.     else:  
183.         y = 7          
184.     del row[‘Label’]  
185.       
186.     # 处理特征  
187.     data = []  
188.     for key in row:  
189.         value = string.atof(row[key])  
190.         data.append(value)  
191.       
192.     p = mat(data)  
193.     HTrain = sig(p, Iw, bias, 1)  
194.     Y = HTrain * beta  
195.       
196.     # 判断  
197.     if argmax(Y) + 1 == y:  
198.         correctTest += 1  
199.     sumTest += 1  
200. print(“测试准确性为：%f” % (correctTest/sumTest))  

# coding:UTF-8

#

OS_ELM

author : zhiyong_will

date : 2015.3.22

#

from future import division
from datetime import datetime
from csv import DictReader
from math import exp
import random
from numpy import *
import string

设置相关参数

trainData = "C:\Users\dell\Desktop\OS-ELM\segment_train.csv"
testData = "C:\Users\dell\Desktop\OS-ELM\segment_test.csv"

隐含神经元的个数

nHiddenNeurons = 180

输入层的神经元个数

nInputNeurons = 19

初始训练集的大小

NO = 280

函数

def sig(tData, Iw, bias, num):
'''
tData:样本矩阵：样本数*特征数
Iw:输入层到第一个隐含层的权重：隐含层神经元数*特整数
bias:偏置1*隐含神经元个数
'''
v = tData * Iw.T #样本数*隐含神经元个数
bias_1 = ones((num, 1)) * bias
v = v + bias_1
H = 1./(1+exp(-v))
return H

导入数据集

firstTrainData = []
firstTrainLable = []

处理训练样本

for t, row in enumerate(DictReader(open(trainData))):
Id = row['Id']
del row['Id']
del row['I0']

data = []
if int(Id) < NO:
    # 处理是否被点击
    if row['Label'] == '1.00000000':
        y = 1
    elif row['Label'] == '2.00000000':
        y = 2
    elif row['Label'] == '3.00000000':
        y = 3
    elif row['Label'] == '4.00000000':
        y = 4
    elif row['Label'] == '5.00000000':
        y = 5
    elif row['Label'] == '6.00000000':
        y = 6
    else:
        y = 7        
    del row['Label']
    firstTrainLable.append(y)
    # 处理特征
    for key in row:
        value = string.atof(row[key])
        #index = int(value + key[1:], 16) % D
        data.append(value)

    firstTrainData.append(data)
    continue
elif int(Id) == NO:#开始训练
    p0 = mat(firstTrainData)
    T0 = zeros((NO, 7))
    #处理样本标签
    for i in xrange(0, NO): 
        a = firstTrainLable[i]
        T0[i][a-1] = 1

    T0 = T0 * 2 - 1
    Iw = mat(random.rand(nHiddenNeurons, nInputNeurons) * 2 - 1)#随机生成区间-1,1之间的随机矩阵
    bias = mat(random.rand(1, nHiddenNeurons))
    H0 = sig(p0, Iw, bias, NO)#样本数*隐含神经元个数
    M = (H0.T * H0).I
    beta = M * H0.T * T0
else:#训练剩余的样本,每次训练一条样本
    # 处理label
    if row['Label'] == '1.00000000':
        y = 1
    elif row['Label'] == '2.00000000':
        y = 2
    elif row['Label'] == '3.00000000':
        y = 3
    elif row['Label'] == '4.00000000':
        y = 4
    elif row['Label'] == '5.00000000':
        y = 5
    elif row['Label'] == '6.00000000':
        y = 6
    else:
        y = 7        
    del row['Label']
    Tn = zeros((1, 7))
    #处理样本标签
    b = y
    Tn[0][b-1] = 1
    Tn = Tn * 2 - 1
    # 处理特征
    data = []
    for key in row:
        value = string.atof(row[key])
        data.append(value)
    pn = mat(data)
    H = sig(pn, Iw, bias, 1)
    M = M - M * H.T * (eye(1,1) + H * M * H.T).I * H * M
    beta = beta + M * H.T * (Tn - H * beta)

计算训练误差

correct = 0
sum = 0
for t, row in enumerate(DictReader(open(trainData))):
del row[‘Id’]
del row[‘I0’]

# 处理是否被点击
if row['Label'] == '1.00000000':
    y = 1
elif row['Label'] == '2.00000000':
    y = 2
elif row['Label'] == '3.00000000':
    y = 3
elif row['Label'] == '4.00000000':
    y = 4
elif row['Label'] == '5.00000000':
    y = 5
elif row['Label'] == '6.00000000':
    y = 6
else:
    y = 7        
del row['Label']

# 处理特征
data = []
for key in row:
    value = string.atof(row[key])
    data.append(value)

p = mat(data)
HTrain = sig(p, Iw, bias, 1)
Y = HTrain * beta

# 判断
if argmax(Y) + 1 == y:
    correct += 1
sum += 1

print(“训练准确性为：%f” % (correct/sum))

计算测试误差

correctTest = 0
sumTest = 0
for t, row in enumerate(DictReader(open(testData))):
del row[‘Id’]
del row[‘I0’]

# 处理是否被点击
if row['Label'] == '1.00000000':
    y = 1
elif row['Label'] == '2.00000000':
    y = 2
elif row['Label'] == '3.00000000':
    y = 3
elif row['Label'] == '4.00000000':
    y = 4
elif row['Label'] == '5.00000000':
    y = 5
elif row['Label'] == '6.00000000':
    y = 6
else:
    y = 7        
del row['Label']

# 处理特征
data = []
for key in row:
    value = string.atof(row[key])
    data.append(value)

p = mat(data)
HTrain = sig(p, Iw, bias, 1)
Y = HTrain * beta

# 判断
if argmax(Y) + 1 == y:
    correctTest += 1
sumTest += 1

print(“测试准确性为：%f” % (correctTest/sumTest))
实验结果：

参考文章：

A Fast and Accurate Online Sequential Learning Algorithm for Feedforward Networks

        </div>
            </div>