matlab 贝叶斯决策对二维二分类问题的实现

load data1;
right=0;
x=zeros(2,1);
%假设它们都符合二元高斯分布
u1=zeros(2,1);%值为1的类的均值
v1=zeros(2,2);%值为1的类的方差
u2=zeros(2,1);%值为2的类的均值
v2=zeros(2,2);%值为2的类的方差
sum1=0;
sum2=0;
pior1=0;
posterior1=0;
likehood1=0;
pior2=0;
posterior2=0;
likehood2=0;

for i=1:1:2500%求先验
    if(data(3,i)==1)
	   sum1=sum1+1;
	else
       sum2=sum2+1;
	end
end
pior1=sum1/(sum1+sum2);
pior2=sum2/(sum1+sum2);


for m=2501:1:3000  %测试循环

for i=1:1:2500%求均值
    if(data(3,i)==1)
	   u1(1,1)=u1(1,1)+data(1,i);
	   u1(2,1)=u1(2,1)+data(2,i);
	else
	   u2(1,1)=u2(1,1)+data(1,i);
	   u2(2,1)=u2(2,1)+data(2,i);
	end
end
u1=u1/2500;
u2=u2/2500;

for i=1:1:2500%求方差
    if(data(3,i)==1)
	   x(1,1)=data(1,i);
	   x(2,1)=data(2,i);
	   v1=v1+(x-u1)*(x-u1)';
	else
	   x(1,1)=data(1,i);
	   x(2,1)=data(2,i);
	   v2=v2+(x-u2)*(x-u2)';
	end
end
v1=v1/2500;
v2=v2/2500;

%求似然
x(1,1)=data(1,m);
x(2,1)=data(2,m);
likehood1=1/(det(v1)^0.5)*exp(-0.5*(x-u1)'*(v1^-1)*(x-u1));%相同的项不乘了
likehood2=1/(det(v2)^0.5)*exp(-0.5*(x-u2)'*(v2^-1)*(x-u2));

求后验
posterior1=pior1*likehood1;
posterior2=pior2*likehood2;

if(((posterior1>posterior2)&&(data(3,m)==1))||((posterior1<posterior2)&&(data(3,m)==-1)))
   right=right+1;
end
   
end

 

数据下载点击打开链接