-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrainWaldBoostLearner.m
176 lines (163 loc) · 7.07 KB
/
trainWaldBoostLearner.m
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
169
170
171
172
173
174
175
% WaldBoost algorithm
% WaldBoostClassfy Classify the training sample
% searchBestWeakLearner Get the best weak learner
% trainWaldBoostLearner Waldboost training process
% testWaldBoostLearner Waldboost testing process
% testWaldBoost Generate data sample and call Waldboost to train and test
% WaldBoost WaldBoost train and test entry
%
% Input:
% X train set
% integral patch
% cntSamples*768 matrix from cntSamples*16*16*3
% Y correct train set tag
% T training iteration time
% cntSelectFeatures
% 可选参数，需要训练指定数量的特征的数量
% 若输入4个参数，则表示不会依照训练轮数为T的规则
% 此时将一直训练，直至分类器训练出指定的特征
% 在这种情况下 costTime 指提取单个特征所获得的时间
%
% Output：
% Hypothesis Weak learner
% thres sign feature
% -16.4151 1 27
% 0.0073 1 291
% 0.4482 -1 14
% 0.0540 1 315
%
% AlphaT Weight for each weak learner
% trainErrorRate training error. 1*T vector
% costTime cost time over T iteration
%
% call example：
% [Hypothesis,AlphaT,trainErrorRate,OverallErrorRate,costTime,TPRate,FPRate]=trainWaldBoostLearner(X,Y,T)
% [Hypothesis,AlphaT,trainErrorRate,costTime]=trainWaldBoostLearner(X,Y,T,cntSelectFeatures)
%
%
function [Hypothesis,AlphaT,trainErrorRate,OverallErrorRate,costTime,TPRate,FPRate]=trainWaldBoostLearner(X,Y,T,varargin)
error(nargchk(3,4,nargin)); % 必须输入3-4个参数,否则中止程序
iptcheckinput(X,{'numeric'},{'2d','real','nonsparse'}, mfilename,'X',1);
iptcheckinput(Y,{'logical','numeric'},{'row','nonempty','integer'},mfilename, 'Y', 2);
iptcheckinput(T,{'numeric'},{'row','nonempty','integer'},mfilename, 'T',3);
if( length(T) > 1 ) % 指定训练轮数的参数T长度应为1（不能为向量）
error(['T should be a number']);
end
[cntSamples,cntFeatures]=size(X); % cntSamples train set size
inverseControl=0; % loop control
cntSelectFeatures=0; % feature number to iterate
if( nargin>3 ) % 4 input
cntSelectFeatures=varargin{1};
inverseControl=1;
iptcheckinput(cntSelectFeatures,{'numeric'},{'row','nonempty','integer'},mfilename, 'cntSelectFeatures',4);
if( length(cntSelectFeatures) > 1 ) % feature number should not be vector
error(['(cntSelectFeatures) should be a number']);
end
if( cntSelectFeatures>=cntFeatures )
error('(cntSelectFeatures) too big！');
end
end
if( cntSamples~=length(Y) ) % check if match
error('error cntSamples length does not match tag length') ;
end
computeCosttimeFlag=1; % counting flag. count time if equal to 1
if(computeCosttimeFlag==1)
tic
end
X=ceil(X*10000)/10000; % rounding X
positiveCols=find(Y==1); % cols for tag = 1
negativeCols=find(Y==-1); % cols for tag = -1
if(length(positiveCols)==0) % check not equal to 0
error('positive sample size 0.');
end
if(length(negativeCols)==0) % check not equal to 0
error('negative sample size 0.');
end
weight=ones(1,cntSamples); % initialize weight for samples
weight(positiveCols)=1/(2*length(positiveCols));
weight(negativeCols)=1/(2*length(negativeCols));
Hypothesis=zeros(T,20); % initialize output value
AlphaT=zeros(1,T);
trainErrorRate=zeros(1,T);
OverallErrorRate=zeros(1,T);
costTime=zeros(1,T);
TPRate = zeros(1,T);
FPRate = zeros(1,T);
trainOutput=zeros(1,cntSamples); % intermediate vector: store the output from temp learner
h=zeros(1,cntSamples); % intermediate vector: store the output of best weaker learner
Decision=zeros(1,cntSamples);
t=1;
curFeaSize=0; % size of features which have been used for training
featureCascadeNum = 50;
while(true)
minError=cntSamples; % Misclassified initialization：cntSamples
%weight=weight/(sum(weight));% Normalize weight
undecided_idx = find(Decision==0); % 1. remove decided samples
%undecided_idx = [1:cntSamples]; % 2. keep all samples
[t,size(undecided_idx,2)] %debug info
samples_X = X(undecided_idx,:);
samples_Y = Y(undecided_idx);
samples_weight = weight(undecided_idx);
samples_weight=samples_weight/(sum(samples_weight));
%[HypothesisOut predout] = BestAdaBoostWeakLearnerManual(samples_X(:,:),samples_Y,samples_weight);
[HypothesisOut predout] = BestAdaBoostWeakLearner(samples_X(:,:),samples_Y,samples_weight,featureCascadeNum);
lengthofhyp = length(HypothesisOut);
Hypothesis(t,1:lengthofhyp) = HypothesisOut;
%h=WaldBoostWeakLearnerClassfy(samples_X,Hypothesis(t,:)); % classify with best learner
errorRate=sum(samples_weight(find(predout~=samples_Y)));
errorRate % calculate the misclassifed ratio
AlphaT(t)=log10((1-errorRate)/(errorRate+eps)); % calculate the weight
if(errorRate>eps) % weight down the correctly classified samples
samples_weight(find(predout==samples_Y))=samples_weight(find(predout==samples_Y))*(errorRate/(1-errorRate));
end
weight(undecided_idx) = samples_weight(:);
% at t iteration, get the error rate and threshold for current strong learner
%[trainOutput,thresA,thresB]=WaldBoostClassfy(samples_X,samples_Y,Hypothesis,AlphaT,t);
[trainOutput,thresA,thresB]=WaldBoostCl(samples_X,samples_Y,Hypothesis,AlphaT,t,t==T);
idx = find(trainOutput~=0);
newdecided_idx = undecided_idx(idx);
Decision(newdecided_idx) = trainOutput(idx);
[curErrorRate,curOverallErrorRate,curTPRate,curFPRate]=calPredictErrorRate(Y,Decision);
Hypothesis(t,1:2) = [thresA thresB];
%t
%curErrorRate
%size(curErrorRate,2)
%if(size(curErrorRate,2)>1)
% curErrorRate = curErrorRate(1,1);
%end
%whos curErrorRate
trainErrorRate(t) = curErrorRate;
OverallErrorRate(t) = curOverallErrorRate;
TPRate(t) = curTPRate;
FPRate(t) = curFPRate;
if(inverseControl==0) % control if 4 input
if(computeCosttimeFlag==1)
costTime(t)=toc;
end
if(t>=T) % reach required iteration number
break;
end
else % feature control
[SelectFeaNo]=analysisHypothesisFeature(Hypothesis(1:t,:),0);
if( length(SelectFeaNo)>curFeaSize )
curFeaSize=length(SelectFeaNo);
if( computeCosttimeFlag==1 )
costTime(curFeaSize)=toc;
end
end
if( curFeaSize>=cntSelectFeatures )% reach feature size
break;
end
end
t=t+1;
end
if(computeCosttimeFlag==1)
costTime=costTime(find(costTime~=0));% ouput cost time
else
costTime=0;
end
if(t<T)
Hypothesis=Hypothesis(1:t,:); % strong classifier
AlphaT=AlphaT(1:t); % strong classifier weight
trainErrorRate=trainErrorRate(1:t);% training error
end