matlab编写KNN和EM算法

KNN：

function [result,tag]=KNN(X,Y,k) %这里Y为一行多列
[m,n]=size(X);
tag=0;
result=0;

for i=1:m
sum=0;
for j=1:n
if (isnan(Y(j))==0)
sum=sum+(X(i,j)-Y(j))^2;
else
tag=j;
end
end
Z(i,1:2)=[i,sqrt(sum)];
end

Z=sortrows(Z,2);%进行排序
Q=Z(1:k,:);
sum_t=0;
for i=1:k
sum_t=sum_t+Q(i,2);
end

Q(:,2)=Q(:,2)/sum_t;

%计算求值
for i=1:k
result=result+X(Q(i,1),tag)*Q(i,2);
end

endEM：
function [result,tag1]=EM(X,Y,e)
A1=[0.25 0.25 0.25 0.25]; %初始系数
[m,n]=size(X); %获得矩阵大小
tag1=zeros(1,1); %标志
result=zeros(1,1); %

%获取缺失数据的位置
for i=1:n
if isnan(Y(i))==1 %判断是否是缺失数据位置
if tag1(1,1)==0
tag1=i;
else
tag1=[tag1,i];
end
tag=i;

u1=[mean(X(1:40,tag)),mean(X(41:80,tag)),mean(X(81:120,tag)),...
mean(X(121:160,tag))];
b1=[var(X(1:40,tag),1),var(X(41:80,tag),1),...
var(X(81:120,tag),1),var(X(121:160,tag),1)];

A=zeros(1,4);
u=zeros(1,4);
b=zeros(1,4);
count=0;

while (judge(A,A1,u,u1,b,b1,e)<1)
count=count+1;
if count>1000
break;
end
%开始进行计算分类
N_1=zeros(1,1);
N_2=zeros(1,1);
N_3=zeros(1,1);
N_4=zeros(1,1);

for j=1:m
for k=1:4
compare(k)=A1(k)*exp((-1)*((X(j,tag)-u1(k))^2)/2/b1(k))/sqrt(b1(k));
end
% compare
[M I]=max(compare); %获得概率最大值和属于第几类I
switch I
case 1
if N_1(1,1)==0
N_1(1,1)=X(j,tag);
else
N_1=[N_1,X(j,tag)];
end

case 2
if N_2(1,1)==0
N_2(1,1)=X(j,tag);
else
N_2=[N_2,X(j,tag)];
end

case 3
if N_3(1,1)==0
N_3(1,1)=X(j,tag);
else
N_3=[N_3,X(j,tag)];
end

case 4
if N_4(1,1)==0
N_4(1,1)=X(j,tag);
else
N_4=[N_4,X(j,tag)];
end
end
end
% l1=length(N_1)
% l2= length(N_2)
% l3=length(N_3)
% l4= length(N_4)
u=u1; %赋值
b1=b;
A=A1; %接下来计算新值
u1=[mean(N_1),mean(N_2),mean(N_3),mean(N_4)];
b1=[var(N_1,1),var(N_2,1),var(N_3,1),var(N_4,1)];
A1=[length(N_1)/m,length(N_2)/m,length(N_3)/m,length(N_4)/m];
% A1
end
[M,I]=max(A1);
if result(1,1)==0
result=u1(I);
else
result=[result,u1(I)];
end

end

end

endfun:
clc;
clear all;
load('totalPDSdat.mat');

X=totalPDSdat(1:160,:);
for i=161:168
Y(i-160,:)=totalPDSdat(i,:);
if i<=164
Y(i-160,1+i-160)=NaN; %将其视为缺失的数据
A(i-160)=totalPDSdat(i,1+i-160);
else
switch(i)
case 165
Y(i-160,13)=NaN; %将其视为缺失的数据
A(i-160)=totalPDSdat(i,13);

case 166
Y(i-160,20)=NaN; %将其视为缺失的数据
A(i-160)=totalPDSdat(i,20);

case 167
Y(i-160,23)=NaN; %将其视为缺失的数据
A(i-160)=totalPDSdat(i,23);

case 168
Y(i-160,24)=NaN; %将其视为缺失的数据
A(i-160)=totalPDSdat(i,24);
end
end
end

Z=Y;
for i=1:8
[result,tag1]=EM(X,Y(i,:),0.001);
[m,n]=size(result);
for j=1:n
Z(i,tag1(j))=result(j);
B(i)=result(j);
end

[result1,tag]=KNN(X,Y(i,:),10);
C(i)=result1;
end

P=1:8;
scatter(P,A,'r','filled');
hold on;
scatter(P,B,'k','filled');
scatter(P,C,'b','filled');
axis([1,8, 0,210]);
legend('原始数据','EM算法','KNN算法');