checkson.m

多元线性回归：MATLAB源程序多元线性回归

%#									
%#  function [ns]=checkson(newson,pop,chrom,xc,yc,cvgp,maxvar)		
%#									
%#  AIM:	Checks whether the two strings obtained after		
%#		cross-over and mutation obey the constraints		
%#		of the GA.						
%#									
%#  PRINCIPLE:	The children strings must obey a few constraints:	
%#		 - they must have at least 1 variable.			
%#		 - they must have at most maxvar variables.		
%#		 - they must be different from each other, and from	
%#		   any other member of the population.			
%#									
%#  INPUT:								
%#	newson	: (2 x col) the two children strings to test.		
%#	pop	: (chrom x col) the actual population of strings.	
%#	chrom	: the number of strings in the population.		
%#	xc, yc	: (row x col) calibration set, on which the GA is run.	
%#	cvgp	: Number of deletion groups for cross-validation.	
%#	maxvar	: Maximum number of variables for selection.		
%#									
%#  OUTPUT:								
%#	ns	: (0,1,or 2 x col) matrix containing the children	
%#		  strings which satisfy the constraints of the GA.	
%#									
%#  SUBROUTINES:							
%#	stepson.m Performs a stepwise elimination if one string		
%#		  contains more than 'maxvar' variables.		
%#									
%#  AUTHOR:     Delphine Jouan-Rimbaud					
%#              Copyright(c) 1997 for ChemoAC				
%#              FABI, Vrije Universiteit Brussel			
%#              Laarbeeklaan 103 1090 Jette				
%#									
%# VERSION: 1.1 (28/02/1998)						
%#									
%#  TEST:	Frederic Despagne					
%#									

function [ns]=checkson(newson,pop,chrom,xc,yc,cvgp,maxvar);

ns=newson;
if sum(newson(2,:)) > sum(newson(1,:)) 
   ns(1,:)=newson(2,:);
   ns(2,:)=newson(1,:);
end
% the strings are re-arranged, so that the one containing
% the highest number of variables is the first one in ns.

sumns1=sum(ns(1,:));  % Number of variables in the first string of ns
sumns2=sum(ns(2,:));  % Number of variables in the second string of ns
sumtot=sumns1+sumns2; % Total number of variables in the two strings.
		      % According to the requirements of the GA, sumtot should not exceed 2 x maxvar

if sumtot <= 2*maxvar
   if sum(ns(1,:)) > maxvar

   % In the original algorithm of Leardi, if the number of variables in
   % ns(1,:) exceeds maxvar, the string is not evaluated, but is killed.
   % In order not to lose what could be some important information, the
   % following is done:
   % Some variables of ns1 are transferred to ns2. This is possible, as
   % there are fewer than 'maxvar' variables in ns(2,:).
   % This step is a kind of cross-over, and avoids to discard blindly a potentially informative subset.

	nb1=sum(ns(1,:))-maxvar; % Number of variables in excess
	dns=ns(1,:)-ns(2,:);   
	f1=find(dns==1) ; % variables of ns(1,:) which are not selected in ns(2,:)
	lf1=length(f1);
	r1=randperm(lf1);
	f1=f1(r1);
	ns(1,f1(1:nb1))=zeros(1,nb1);
	ns(2,f1(1:nb1))=ones(1,nb1);
	% one randomly eliminates 'nb1' variables from ns(1,:),
	% and places them into ns(2,:).
   end

   ns1=ns(1,:);
   ns2=ns(2,:);

   if ns1==ns2	% The two children strings cannot be equal.
		% If they are, the less informative variable is removed from one string.
		% This is done in STEPSON.M
	[string1]=stepson(ns1,xc,yc,cvgp,1);
	ns1=string1; clear string1
   end
end

if sumtot > 2*maxvar
   % In this case, one cannot transfer all variables in excess
   % from one string to the other one.
   % The less informative variables are removed from the string,
   % until 'maxvar' variables remain.

   ns1=ns(1,:); ns2=ns(2,:);

   if sumns1 > maxvar
	nb_iter=sumns1-maxvar; % Number of variables in excess, which will be removed from ns1.
	[string1]=stepson(ns(1,:),xc,yc,cvgp,nb_iter);
	ns1=string1; clear string1
   end
   if sumns2 > maxvar
	nb_iter=sumns2-maxvar; % Number of variables in excess, which will be removed from ns2.
	[string2]=stepson(ns(2,:),xc,yc,cvgp,nb_iter);
	ns2=string2; clear string2
   end

   if ns1==ns2	% the two children strings cannot be equal.
		% if they are, the less informative variable is removed from one string.
	[string1]=stepson(ns1,xc,yc,cvgp,1);
	ns1=string1; clear string1
   end
end

if ns1==zeros(1,length(ns1)), ns1=[]; end
if ns2==zeros(1,length(ns1)), ns2=[]; end
% If one of the two strings is empty, it is replaced by the empty vector.


% Now, it is checked that none of the "surviving" string(s) is (are) similar to one string already in the population.
% Otherwise, this child string is discarded.

if ns1~=[],
   j=0;
   while j<chrom
	j=j+1;
	if pop(j,:)==ns1
	   ns1=[];
	   j=chrom+1;
	end
   end
end

if ns2~=[],
   j=0;
   while j<chrom
	j=j+1;
	if pop(j,:)==ns2
	   ns2=[];
	   j=chrom+1;
	end
   end
end

ns=[ns1;ns2];
% Contains 0, 1, or 2 strings, which obey the requirements of the GA.