da_plsgo.m

王小平《遗传算法——理论、应用与软件实现》随书光盘

%
% mv_pcrgo.m
%
% Gets all the parameters for the principal
% components regression subprogramme
%

%
% Separate input and output variables
%

y=data(:,output_var & include_var);
x=pcincl(data,include_var,output_var);

%
% Scale the data
%

[y my sdy]=mcen(y);
[x mx sdx]=mcen(x);

%
% Split the data correctly
%

%
% Check radio button 4 for the percentage partition
%

if get(pcr_rad4,'Value') == 1
	[tx ty vx vy]=psplit(x,y,get(pcr_rad4sl,'Value'));
end

%
% Check radio button 5 for the random partition
%

if get(pcr_rad5,'Value') == 1
	disp('Bollocks')
	[tx ty vx vy]=rsplit(x,y,get(pcr_rad5sl,'Value'));
end

%
% Check radio button 6 for the block splitting
%

if get(pcr_rad6,'Value') == 1
	[tx ty vx vy]=bsplit(x,y,ceil(get(pcr_rad6sl,'Value')));
end

%
% Check radio button 7 for alternate split
%

if get(pcr_rad7,'Value') == 1
	[tx ty vx vy]=asplit(x,y);
end


%
% Check radio button 1 for include number of latent variables
%

if get(pcr_rad1,'Value') == 1
	%
	% Determine number of latent varibles to include
	%
	lv=ceil(get(sl2,'Value'));

	%
	% Calculate scores etc
	%
	[P Q W T U b ssq]=pls(tx,ty,lv);

	%
	% Calculate predicted y based on this information
	%
	py=plspred(vx,b,P,Q,W,lv);

	%
	% Convert the pls model to a regression vector
	% which gives the impact of each variable
	%
	impact=conpred1(b,W,P,Q,lv);

	%
	% Delete the pls model to save memory
	%
	clear P Q W b ssq
end

%
% Include the optimum number of latent variables
%
if get(pcr_rad3,'Value') == 1
	%
	% maximum number of latent variables = number of variables
	% included in the analysis -1 (output variables)
	%
	max_lv=sum(include_var)-sum(output_var);

	%
	% Build a pls model using this number of lv's
	%
	[P Q W T U b ssq]=pls(tx,ty,max_lv);

	%
	% Progressively increase the number of lv's to
	% find the minimum sse
	%

	max_err=1e100;
	
	for i = 1 : max_lv

		py=plspred(vx,b,P,Q,W,i);
		e=sse(py,vy);	
		if e < max_err
			max_err=e;
			best_lv=i;
			best_py=py;
			acty=plspred(tx,b,P,Q,W,i);
			impact=conpred1(b,W,P,Q,i);
		end
	end
	py=best_py;
	lv=best_lv;
	
	%
	% Now clear the model out of memory
	%
	clear P Q b ssq best_lv best_py max_err max_lv e
	acty=rescale(acty,my,sdy);
end





%
% Rescale the data
%
%tx=rescale(tx,mx,sdx);
ty=rescale(ty,my,sdy);

%vx=rescale(vx,mx,sdx);
vy=rescale(vy,my,sdy);

py=rescale(py,my,sdy);

%
% Print the results in the little blue box at the top of the screen
%


[D L]=size(vy);
e=((sse(vy(:,1),py(:,1)).^0.5)/D)/my(1);
c=corrcoef(vy(:,1),py(:,1));
c=c(2,1);
regv=impact;
impact=impact';

text1=uicontrol(w1,...
	'Style','text',...
	'Position',[275 385 130 15],...
	'BackGroundColor',[0 0 1],...
	'ForeGroundColor',[1 1 1],...
	'String',['Error =  ' num2str(e)]);
text2=uicontrol(w1,...
	'Style','text',...
	'Position',[275 365 145 15],...
	'BackGroundColor',[0 0 1],...
	'ForeGroundColor',[1 1 1],...
	'String',['Latent variables =  ' num2str(lv)]);
text3=uicontrol(w1,...
	'Style','text',...
	'Position',[410 385 130 15],...
	'BackGroundColor',[0 0 1],...
	'ForeGroundColor',[1 1 1],...
	'String',['Correlation =  ' num2str(c)]);

%
% Save results button
%
sl1_value=1;
pca_but2=uicontrol(w1,...
	'Style','push',...
	'Position',[380 10 100 20],...
	'String','More detail',...
	'CallBack',[...
		'py=py(:,sl1_value);',...
		'vy=vy(:,sl1_value);',...
		'impact=impact(sl1_value,:);',...
		'da_det1;']);
clear x y regv


if sum(output_var) > 1
	set(text4,'visible','on');
	set(text5,'visible','on');
	set(sl1,'visible','on','enable','on');
	sl1_value=1;
	old_sl1_value=1;
	set(sl1,'value',1);
	set(text4,'string','Viewing output variable 1');
end

%
% Button to plot U v T
%
%but4=uicontrol(w1,...
%	'Style','push',...
%	'Position',[500 10 100 20],...
%	'String','U v T plot',...
%	'CallBack','da_plsut;');

pca_but3=uicontrol(w1,...
	'style','push',...
	'position',[500 10 100 20],...
	'string','Save model',...
	'callback',[...
		'[D L]=size(tx);',...
		'out=svlin2('''',include_var,sdx,mx,sdy,my,impact,ones(L,1),zeros(L,1))']);