pls程序0.txt

kennard-stone选样本算法。。。PLS的迭代算法程序

function [B,p,w,b]=pls1(x,y,A); 

%[p,w,b]=pls1(x,y,maxnum) is a program by using pls1 algorithm to build a PLS model. 
%x is a n*m matrix,where m is wavelength number and n sample number. 
%y is a n*1 vector.   
% A is a input parameter stated maximum of principle component number. 
%output p is loadings, w weights and b is the regression coefficient. 

% example for the application of the program  
% 
%[p,w,b]=pls1(x,y,10); 
%for i=1:10    % with different latent variable number 
%   beta=w(:,1:i)/(p(:,1:i)'*w(:,1:i))*b(1:i); 
%   ype=xp*beta;   % xp is a spectrum or spectra (a row vector or a matrix) to be used to predict 
%   SEP=[SEP sqrt((yp-ype)'*(yp-ype)/(length(yp)))];% calculate the standard error of prediction 
%end; 

% By Yiping Du, 2002,10,30, modeified in 2003,06,17  

[n,m]=size(x);% 
[nc,mc]=size(y); 
if n~=nc 
   errordlg('error input of x and y','pls'); 
   return; 
end; 
n_p=min([m,n,A]); 
t=[];p=[];b=[];w=[];resi_y=[];resi_x=[]; 
for i=1:n_p 
   t=y;%one column with n elements 
   w=t'*x/(t'*t);%one row with m elements, w is mid var. of p 
   w=w/norm(w,2);%norm(w,2)=sqrt(w*w') 
   t=x*w'/(w*w');% one column with n elements 
   p=t'*x/(t'*t);  
   b=y'*t/(t'*t);%b is actually the q in models usually used. 
   pp(i,:)=p; 
   tt(:,i)=t; 
   ww(i,:)=w; 
   bb(i)=b; 
   x=x-t*p; 
   y=y-b*t;   
end; 
p=pp'; 
w=ww'; 
b=bb'; 
B=w(:,1:A)/(p(:,1:A)'*w(:,1:A))*b(1:A); 
%resi_x=x; 
%resi_y=y; 

--