-
Notifications
You must be signed in to change notification settings - Fork 37
/
6.AdaBoost.py
168 lines (133 loc) · 8.2 KB
/
6.AdaBoost.py
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
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
#!/usr/bin/env python3
# -*-coding: utf-8-*-
# Author : LiangjunFeng
# Blog : http://my.csdn.net/Liangjun_Feng
# GitHub : https://www.github.com/LiangjunFeng
# File : AdaBoost.py
# Date : 2017/09/17 11:12
# Version: 0.1
# Description: AdaBoost algorithm ,ensemble learning
import numpy
import math
from pandas import read_csv
def loadData(): #load the CSV data file
dataSet = read_csv('/Users/zhuxiaoxiansheng/Desktop/data_set/pima-indians-diabetes/pima-indians-diabetes.csv',header = None)
return dataSet
#data preprocess,split the atributes data and the label data,
#meanwhile filling the lack data,tranform the form of the data to matrix
def dataPrepreocess(dataSet):
dataSet[[0,1,2,3,4,5,6,7]] = dataSet[[0,1,2,3,4,5,6,7]].replace(0,numpy.NaN)
data = dataSet[[0,1,2,3,4,5,6,7]]
data.fillna(data.median(),inplace = True) #filling the lack with median value
label = dataSet[8]
data = numpy.matrix(data.as_matrix(columns=None)) #transfer to matrix
label = numpy.matrix(label.as_matrix(columns=None)) #transfer to matrix
return data,label.T
def dataSplit(data,label): #split the atributes data and the label data
for i in range(len(label)):
if label[i,0] == 0:
label[i,0] = -1
mid = int(len(data)*0.68)
trainData = data[0:mid] #set the top 60% as train data
testData = data[mid:] #set the left as tet data
trainLabel = label[0:mid,0]
testLabel = label[mid:,0]
return trainData,trainLabel,testData,testLabel #return train data,train label,test data,test label
class decisionRoot: #use decision tree root as base classifier
def __init__(self):
self._rootAttribute = 0 #the tree root's classification attribute
self._thresholdValue = 0 #the threshold value of the tree root's classification attribute
self._minError = 0 #the classifiers' error of classification
self._operator = None #the classifier operator，the value could be ‘le’ or ‘gt’
self._estimateLabel = None #the estimated label or the train set
@classmethod
def classify(cls,data,operator,thresholdValue): #have known the root's classification attribute,threshold value and classifier operator,classifiy the data to two classes
estimateLabel = numpy.ones((data.shape[0],1))
if operator == 'le':
estimateLabel[data <= thresholdValue] = -1 #if the operator is 'le' and if the attribute value less or equals the threshold value,set the label with -1,the else remain be 1
else:
estimateLabel[data > thresholdValue] = -1 #if the operator is 'gt' and if the attribute value greater than the threshold value,set the label with -1,the else remain be 1
return estimateLabel
def buildTree(self,data,label,distribute): #derive the optimal Tree root based on the train data and distribute
samples,attributes = data.shape
estimateLabel = numpy.matrix(numpy.zeros((samples,1)))
minError = numpy.inf #set the inital value of the minimal error as infinity
stepNumber = 10.0 #In order to determine the optimal classification threshold of attributes, the sample attributes are traversed from the minimum to the maximum, and the number of steps is 10
for rootAttribute in range(attributes):
attributeMin = data[:,rootAttribute].min()
attributeMax = data[:,rootAttribute].max()
stepSize = (attributeMax - attributeMin)/stepNumber #get the step length
for j in range(int(stepNumber+1)): #Traversal attributes, in order to obtain the optimal classification attributes
for operator in ['le','gt']: #Select the operator to get the best operater
thresholdValue = attributeMin + j*stepSize #the threshold value of the tree root's classification attribute
estimateLabel = decisionRoot.classify(data[:,rootAttribute],operator,thresholdValue) #the label that estimated by the classifier
errLabel = numpy.matrix(numpy.ones((samples,1)))
errLabel[estimateLabel == label] = 0
error = float(distribute*errLabel) #calculate the classification error
if error < minError:
minError = error #record parameter of the minimum error
self._minError = minError
self._rootAttribute = rootAttribute
self._thresholdValue = thresholdValue
self._operator = operator
self._estimateLabel = estimateLabel.copy()
def predict(self,data): #Input data, prediction results
samples = data.shape[0]
result = numpy.matrix(decisionRoot.classify(data[:,self._rootAttribute],self._operator,self._thresholdValue))
return result
class AdaBoost:
def __init__(self):
self._rootNumber = 0 #the root's number that need to train
self._decisionRootArray = [] #store the root that have been trained
self._alphaList = [] #record the alpha value of every root
@classmethod
def exp(cls,vector): #using exp function process all the items in the vector
for i in range(len(vector)):
vector[i,0] = math.exp(vector[i,0])
return vector
def trainRoot(self,data,label,rootNumber = 45): #train root,and store them
samples,attributes = data.shape
self._rootNumber = rootNumber
distribute = numpy.matrix(numpy.ones((1,samples))/float(samples))
for i in range(self._rootNumber): #numbers of roots be trained
root = decisionRoot()
root.buildTree(data,label,distribute) #make and set up a root
alpha = 0.5 * math.log((1.0 - root._minError)/max(root._minError,1e-16))
self._alphaList.append(alpha)
expValue = AdaBoost.exp(numpy.multiply(-1*alpha*label,root._estimateLabel))
distribute = numpy.multiply(distribute,expValue.T)
distribute = distribute/float(distribute.sum()) #refresh the distribute
self._decisionRootArray.append(root)
def vote(self,predictMatrix): #using Voting rules determines classification results
samples = predictMatrix[0].shape[0]
result = []
for i in range(samples):
sign = 0
for j in range(self._rootNumber):
sign += predictMatrix[j][i,0] * self._alphaList[j]
if sign >= 0:
result.append(1)
else:
result.append(-1) #record the result of the voting
return result
def predict(self,data): #Integrated voting results are based on many well trained classifiers
predictMatrix = []
for i in range(self._rootNumber): #Each classifier is traversed to obtain a classification value
root = self._decisionRootArray[i]
result = root.predict(data)
predictMatrix.append(result)
return numpy.matrix(self.vote(predictMatrix)).T #Returns the voting results for each classifier
def assess(predictResult,actualLabel): #Evaluate classifier accuracy
count = 0
for i in range(predictResult.shape[0]):
if predictResult[i,0] == actualLabel[i,0]:
count += 1
return count/ float(predictResult.shape[0])
if __name__ == '__main__':
dataSet = loadData()
data,label = dataPrepreocess(dataSet)
trainData,trainLabel,testData,testLabel = dataSplit(data,label) #Training set and test set are obtained
Ada = AdaBoost()
Ada.trainRoot(trainData,trainLabel)
result = Ada.predict(testData)
print(assess(result,testLabel))