以下是我的Bayes分类器设计的思路，程序原理，运行结果：
设计思路：
基于最小错误率的Bayes分类器
基于最小错误率的Bayes公式
基于最小风险的Bayes分类器
基于最小风险的Bayes公式
程序原理：
最小错误率的多二维特征训练函数
if isempty(get(handles.MaleError_edit,'String')) || isempty(get(handles.FemaleError_edit,'String'))
errordlg('Please input the parameter of Male and Female!','Parameter Error');
return;
end
%获取先验概率
ErrorHeightButton_Status = get(handles.ErrorHeight_radiobutton,'Value');
ErrorWeightButton_Status = get(handles.ErrorWeight_radiobutton,'Value');
if isequal(ErrorHeightButton_Status,0) & isequal(ErrorWeightButton_Status,0)
errordlg('Please Select the option for Training!','Parameter Error');
return;
else
%根据所选择的特征来进行训练
if isequal(ErrorHeightButton_Status,1)
if isequal(ErrorWeightButton_Status,0)
ErrorTrainParameter = 1;
else
ErrorTrainParameter = 3;
end
else
ErrorTrainParameter = 2;
end
end
% Read Files
counter = 1;
while feof(testfile) == 0
%get the data by lines
str = fgetl(testfile);
[str1 str2 str3] = strread(str,'%s %s %s','delimiter',' ');
Test_File(counter) = [str1 str2 str3];
counter = counter + 1;
end
% %get the information of the trained samples
female = handles.female;
male = handles.male;
%test as the matrix of the testfile
test = dlmread(Filename);
[length, width] = size(test);
male_parameter = str2num(get(handles.MaleError_edit,'String'));
female_parameter = str2num(get(handles.FemaleError_edit,'String'));
error1 = 0;
error2 = 0;
%get the mu and sigma of normal distribution
if isequal(ErrorTrainParameter,1)
%Trained by the Parameter of Height
%正态分布估计值函数normfit
[favg, fsig] = normfit(female(:,1));
[mavg, msig] = normfit(male(:,1));
for i=1:length
%正态分布概率值函数normpdf，按照先前的最小错误率公式计算后验概率
Px=normpdf(test(i,1),favg,fsig) * female_parameter + normpdf(test(i,1),mavg,msig) * male_parameter;
P_male=normpdf(test(i,1),mavg,msig) * male_parameter / Px;
P_female=normpdf(test(i,1),favg,fsig) * female_parameter / Px;
if P_male > P_female
if test(i,3) == 2
%错误率计算
error1 = error1 + 1;
test(i,3) = 21;
end
else
if test(i,3) == 1
%错误率计算
error2 = error2 + 1;
test(i,3) = 12;
end
end
end
else
if isequal(ErrorTrainParameter,2)
%Trained by the Parameter of Weight
[favg, fsig] = normfit(female(:,2));
[mavg, msig] = normfit(male(:,2));
[length, width] = size(test);
for i = 1:length
Px = normpdf(test(i,2),favg,fsig) * female_parameter + normpdf(test(i,2),mavg,msig) * male_parameter;
P_male = normpdf(test(i,2),mavg,msig) * male_parameter / Px;
P_female = normpdf(test(i,2),favg,fsig) * female_parameter / Px;
if P_male > P_female
if test(i,3) == 2
error1 = error1 + 1;
test(i,3) = 21;
end
else
if test(i,3) == 1
error2 = error2 + 1;
test(i,3) = 12;
end
end
end
else
if isequal(ErrorTrainParameter,3)
%Trained by the Parameter of Height and Weight
female = female(:,1:2);
male = male(:,1:2);
favg = mean(female);
fsig = cov(female);
mavg = mean(male);
msig = cov(male);
if isequal(get(handles.Irrelated_radiobutton,'Value'),1)
msig(1,2) = 0;
msig(2,1) = 0;
fsig(1,2) = 0;
fsig(2,1) = 0;
end
for i = 1:length
Px = mvnpdf(test(i,1:2),favg,fsig)*female_parameter + mvnpdf(test(i,1:2),mavg,msig)*male_parameter;
P_male = mvnpdf(test(i,1:2),mavg,msig) * male_parameter / Px;
P_female = mvnpdf(test(i,1:2),favg,fsig) * female_parameter / Px;
if P_male > P_female
if test(i,3) == 2
error1 = error1 + 1;
test(i,3) = 21;
end
else
if test(i,3) == 1
error2 = error2 + 1;
test(i,3) = 12;
end
end
end
end
end
end
%平均错误率计算
errorratio = (error1+error2)/length;
set(handles.filename_text,'string',Filename);
set(handles.ErrorResult_text,'string',num2str(errorratio));
Result = mat2cell(test(:,3),length);
set(handles.Test_edit,'string',Test_File);
set(handles.Result_edit,'string',Result);
handles.Result_edit = handles.Test_edit;
fclose(testfile);
guidata(hObject,handles);
基于最小风险的多维特征训练函数
if isempty(get(handles.MaleRisk_edit,'String')) || isempty(get(handles.FemaleRisk_edit,'String'))
errordlg('Please input the parameter of Male and Female!','Parameter Error');
return;
end
%读取先验概率
RiskHeightButton_Status = get(handles.RiskHeight_radiobutton,'Value');
RiskWeightButton_Status = get(handles.RiskWeight_radiobutton,'Value');
if isequal(RiskHeightButton_Status,0) & isequal(RiskWeightButton_Status,0)
errordlg('Please Select the option for Training!','Parameter Error');
return;
else
%多维特征的选择
if isequal(RiskHeightButton_Status,1)
if isequal(RiskWeightButton_Status,0)
RiskTrainParameter = 1;
else
RiskTrainParameter = 3;
end
else
RiskTrainParameter = 2;
end
end
female = handles.female;
male = handles.male;
test = dlmread(Filename);
%设置风险矩阵的初始值
Decision = [0 5;1 0];
[length, width] = size(test);
male_parameter = str2num(get(handles.MaleRisk_edit,'String'));
female_parameter = str2num(get(handles.FemaleRisk_edit,'String'));
error1 = 0;
error2 = 0;
% get the parameters of the normal distribution
if isequal(RiskTrainParameter,1)
% Trained by the parameter of Height
[favg, fsig] = normfit(female(:,1));
[mavg, msig] = normfit(male(:,1));
for i = 1:length
%按照最小风险的公式计算
Px = normpdf(test(i,1),favg,fsig) * female_parameter + normpdf(test(i,1),mavg,msig) * male_parameter;
P_male = normpdf(test(i,1),mavg,msig) * male_parameter / Px;
P_female = normpdf(test(i,1),favg,fsig) * female_parameter / Px;
Pisk_male = Decision(1,1) * P_male + Decision(1,2) * P_female;
Pisk_female = Decision(2,1) * P_male + Decision(2,2) * P_female;
if Pisk_male > Pisk_female
if test(i,3) == 1
%错误率计算
error1 = error1 + 1;
test(i,3) = 12;
end
else
if test(i,3) == 2
%错误率计算
error2 = error2 + 1;
test(i,3) = 21;
end
end
end
else
if isequal(RiskTrainParameter,2)
% Trained by the parameter of Weight
[favg, fsig] = normfit(female(:,2));
[mavg, msig] = normfit(male(:,2));
for i = 1:length
Px = normpdf(test(i,2),favg,fsig) * female_parameter + normpdf(test(i,2),mavg,msig) * male_parameter;
P_male = normpdf(test(i,2),mavg,msig) * male_parameter / Px;
P_female = normpdf(test(i,2),favg,fsig) * female_parameter / Px;
Pisk_male = Decision(1,1)*P_male+Decision(1,2) * P_female;
Pisk_female = Decision(2,1)*P_male+Decision(2,2) * P_female;
if Pisk_male > Pisk_female
if test(i,3) == 1
error1 = error1 + 1;
test(i,3) = 12;
end
else
if test(i,3)==2
error2 = error2 + 1;
test(i,3) = 21;
end
end
end
else
if isequal(RiskTrainParameter,3)
% Trained by the parameter of Height and Weight
female = female(:,1:2);
male = male(:,1:2);
favg = mean(female);
fsig = cov(female);
mavg = mean(male);
msig = cov(male);
if isequal(get(handles.Irrelated_radiobutton,'Value'),1)
msig(1,2) = 0;
msig(2,1) = 0;
fsig(1,2) = 0;
fsig(2,1) = 0;
end
for i = 1:length
Px = mvnpdf(test(i,1:2),favg,fsig) * female_parameter + mvnpdf(test(i,1:2),mavg,msig) * male_parameter;
P_male = mvnpdf(test(i,1:2),mavg,msig) * male_parameter / Px;
P_female = mvnpdf(test(i,1:2),favg,fsig) * female_parameter / Px;
Risk_male = Decision(1,1)*P_male+Decision(1,2) * P_female;
Risk_female = Decision(2,1)*P_male+Decision(2,2) * P_female;
if Risk_male > Risk_female
if test(i,3) == 1
error1 = error1 + 1;
test(i,3) = 12;
end
else
if test(i,3) == 2
error2 = error2 + 1;
test(i,3) = 21;
end
end
end
end
end
end
%平均错误率计算
errorratio = (error1+error2)/length;
set(handles.filename_text,'string',Filename);
set(handles.ErrorResult_text,'string',num2str(errorratio));
set(handles.Test_edit,'string',Test_File);
Result = mat2cell(test(:,3),length);
set(handles.Result_edit,'string',Result);
handles.Result_edit = handles.Test_edit;
fclose(testfile);
guidata(hObject,handles);
运行结果：
基于最小错误率的结果
表4.1条件：相关，训练样本为Test1.txt。
男女概率 身高 体重 两者
0.9 Vs 0.1 0.11429 0.2 0.11429
0.5 Vs 0.5 0.057143 0.057143 0.028571
0.75 Vs 0.25 0.085714 0.057143 0.085714
表4.2条件：相关，训练样本为Test2.txt。
男女概率 身高 体重 两者
0.9 Vs 0.1 0.056667 0.086667 0.056667
0.5 Vs 0.5 0.09 0.14667 0.10667
0.75 Vs 0.25 0.066667 0.11 0.046667
表4.3条件：不相关，训练样本为Test1.txt。
男女概率 身高 体重 两者
0.9 Vs 0.1 0.11429 0.2 0.085714
0.5 Vs 0.5 0.057143 0.057143 0.028571
0.75 Vs 0.25 0.085714 0.057143 0.028571
表4.4条件：不相关，训练样本为Test2.txt。
男女概率 身高 体重 两者
0.9 Vs 0.1 0.056667 0.086667 0.053333
0.5 Vs 0.5 0.09 0.14667 0.096667
0.75 Vs 0.25 0.066667 0.11 0.063333
基于最小风险的结果
表4.5条件：相关，训练样本为Test1.txt
男女概率 身高 体重 两者
0.9 Vs 0.1 0.085714 0.028571 0.028571
0.5 Vs 0.5 0.14286 0.17143 0.14286
0.75 Vs 0.25 0.085714 0.11429 0.028571
表4.6条件：相关，训练样本为Test2.txt
男女概率 身高 体重 两者
0.9 Vs 0.1 0.063333 0.12 0.08
0.5 Vs 0.5 0.34667 0.29667 0.26333
0.75 Vs 0.25 0.10667 0.22333 0.14333
表4.7条件：不相关，训练样本为Test1.txt
男女概率 身高 体重 两者
0.9 Vs 0.1 0.085714 0.028571 0.028571
0.5 Vs 0.5 0.14286 0.17143 0.16333
0.75 Vs 0.25 0.085714 0.11429 0
表4.8条件：不相关，训练样本为Test2.txt
男女概率 身高 体重 两者
0.9 Vs 0.1 0.063333 0.12 0.076667
0.5 Vs 0.5 0.34667 0.29667 0.16333
0.75 Vs 0.25 0.10667 0.22333 0.11667
总结：
个人的第一次matlab程序，界面设计上还有诸多问题，有些显示不正常，对matlab较为生疏，而作为如今在 模拟仿真方面有绝对优势的matlab，日后必然和我的编程有诸多相关，需要尽快学起。
对于本次编程，老师也说过现在给我们做的一些程序算法都是最基础的，强度还是较低，自己在算法理解和函数调用方面感觉有较大的缺陷，对于matlab的函数包括基本语法还不熟练，种种不完善的地方还有待日后加强。
而本次程序的结果分析报告则已连通程序等内容一同上交，这里的文字只是作为个人总结所用。
尽力而为，做好每次的matlab编程。
抱歉各位，格式对齐我已经设置过，但是没有找到正确的方法，故而委屈各位
