parzen_hypercube.m

用mushrooms数据对模式识别课程讲述的各种模式分类方法[线性分类

clc;
clear;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%读取数据，取16个特征
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
samples = textread('data2000.txt');
samples = samples(:,[1:6,9:15,19:22]);  %17列 第1列标号，16列特征

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%对样本进行归一化处理
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
[ms ns]=size(samples);
TMax=max(samples);
TMin=min(samples);
% 第一列是样本标签，从第二列开始归一化
for i=2:ns
    samples(:,i)=(samples(:,i)-TMin(i))/(TMax(i)-TMin(i));
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%将样本分为测试样本，第一类训练样本，第二类训练样本
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
p = randperm(2000);%对1：2000的整数随机排序
experiment_test=samples(p(1:500),:);%测试样本
exper_test=experiment_test(:,2:ns);%测试样本,不带标签
experiment_train=samples(p(501:2000),:);%训练样本
index1=find(experiment_train(:,1)==1);%找到训练样本中第一类的行号
index2=find(experiment_train(:,1)==2);%找到训练样本中第二类的行号
exper_train_class1=experiment_train(index1,2:ns);%训练样本里属于第一类的样本,不带标签
exper_train_class2=experiment_train(index2,2:ns);%训练样本里属于第二类的样本,不带标签

[m n]=size(exper_test);
[m1 n1]=size(exper_train_class1);
[m2 n2]=size(exper_train_class2);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%计算类条件概率密度
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%计算P(x/wi)=k/(nV)
%取边长hn=1的d=10维超立方体
%首先计算落入每个待测样本点的超立方体的点的个数
hn=1;
V=hn^n;
K1=zeros(m,1);%记录第一类训练样本落入待测样本的超立方体的点的个数
K2=zeros(m,1);%记录第二类训练样本落入待测样本的超立方体的点的个数
R1=zeros(m,1);%记录待测样本在第一类的测试结果
R2=zeros(m,1);%记录待测样本在第二类的测试结果
Result=zeros(m,1);%记录最后分类结果
%先算第一类
for i=1:m%对每个待测样本
    for j=1:m1%检测每个训练样本
        flag=0;
        for k=1:n%检查每个分量abs(xi)<(hn/2)，如果都小于则待测样本落入超立方体
            if(abs(exper_test(i,k)-exper_train_class1(j,k))<(hn/2))
                flag=flag+1;
            end
        end
        if flag==n
                K1(i,1)=K1(i,1)+1;
        end
    end
    R1(i,1)=K1(i,1)/(n*V);
end
%再算第二类
for i=1:m%对每个待测样本
    for j=1:m2%检测每个训练样本
        flag=0;
        for k=1:n%检查每个分量abs(xi)<(hn/2)，如果都小于则待测样本落入超立方体
            if(abs(exper_test(i,k)-exper_train_class2(j,k))<(hn/2))
                flag=flag+1;
            end
        end
        if flag==n
                K2(i,1)=K2(i,1)+1;
        end
    end
    R2(i,1)=K2(i,1)/(n*V);
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%画类条件概率密度图，并分类
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%绘图
plot(R1,'-r','LineWidth',2);
hold on;
plot(R2,'-b','LineWidth',2);
xlabel ('待测样本'); 
ylabel ('类条件概率密度 P（x|wi）');
title ('类条件概率密度图');
%分类
for i=1:m
    if R1(i,1)>R2(i,1)
        Result(i,1)=1;
    else
        Result(i,1)=2;
    end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%分析结果
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
[correct,error,ROC] = analyse_result(experiment_test,Result);