WLETSVR

Ya-Fen Ye, Lan Bai, Yuan-Hai Shao, N.-Y. Deng, Zhen Wang. An efficient weighted twin support vector machine for regression. submitted,(2013).

function PredictY= WLETSVR(TestX,DataTrain,FunPara) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % WLETSVR: An efficient weighted twin support vector machine for regression % PredictY= WLETSVR(TestX,DataTrain,FunPara) % % Input: % TestX - Test Data matrix. Each row vector of fea is a data point. % DataTrain - stuct value in Matlab-----Training data. % % % FunPara - Struct value in Matlab % FunPara.p1: [0,inf] Paramter to tune the weight; % FunPara.p2: [0,inf] Paramter to tune the weight; % FunPara.p3: [0,inf] Paramter to tune the weight; % FunPara.p4: [0,inf] Paramter to tune the weight; % FunPara.p5: [0,inf] Paramter to tune the weight; % FunPara.p6: [0,inf] Paramter to tune the weight; % FunPara.p7: [0,inf] Paramter to tune the weight; % FunPara.p8: [0,inf] Paramter to tune the weight; % FunPara.p9: [0,inf] Paramter to tune the weight; % FunPara.p10: [0,inf] Paramter to tune the weight; % % kerfPara: kernel parameters. See kernelfun.m; % % Output: % Predict_Y - Predict value of the TestX. % % Examples: % load SincTN1.mat X Y TestX TestY % DataTrain.X = X; % DataTrain.Y = Y; % FunPara.p1=2^(-1); % FunPara.p2=2^(-1); % FunPara.p3=0.01; % FunPara.p4=0.01; % FunPara.p5=0.1; % FunPara.p6=0.1; % FunPara.p7=100; % FunPara.p8=100; % FunPara.p9=0.005; % FunPara.p10=0.005; % FunPara.kerfPara.type = 'rbf'; % FunPara.kerfPara.pars = 3; % PredictY= WLETSVR(TestX,DataTrain,FunPara); % % % % Version 1.0 --Jul/2014 % % Written by Ya-Fen Ye (yafenye@163.com) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Initailization %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %tic; c1= FunPara.p1; c2= FunPara.p2; v1= FunPara.p3; v2= FunPara.p4; eps1 = FunPara.p5; eps2 = FunPara.p6; itmax1=FunPara.p7; itmax2=FunPara.p8; tol1=FunPara.p9; tol2=FunPara.p10; kerfPara = FunPara.kerfPara; m = size(DataTrain.X,1); n = size(DataTrain.X,2); e = ones(m,1); beta1=1.6/c1; beta2=1.6/c2; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Compute Kernel %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% kerfPara = FunPara.kerfPara; if strcmp(kerfPara.type,'lin') G = [DataTrain.X e]; else G = [kernelfun(DataTrain.X,kerfPara) e]; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Training data %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if strcmp(kerfPara.type,'lin') Q1=eye(m,m)/c1+G*((G'*G+v1*eye(n+1,n+1))\G'); Q2=eye(m,m)/c2+G*((G'*G+v2*eye(n+1,n+1))\G'); p1=(eye(m,m)-G*((G'*G+v1*eye(n+1,n+1))\G'))*DataTrain.Y+eps1*e; p2=(-eye(m,m)+G*((G'*G+v2*eye(n+1,n+1))\G'))*DataTrain.Y+eps2*e; alpha=inv(Q1)*e;oldalpha=alpha+1;it1=0; gamma=(Q2)\e;oldgamma=gamma+1;it2=0; while it1tol1 z1= Q1*alpha+p1-beta1*alpha; pl=(abs(z1)+z1)/2-p1; oldalpha=alpha; alpha=Q1\pl; it1=it1+1; end; while it2tol2 z2= Q2*gamma+p2-beta2*gamma; p2=(abs(z2)+z2)/2-p2; oldgamma=gamma; gamma=Q2\p2; it2=it2+1; end; u1=(G'*G+v1*eye(n+1,n+1))/G'*(DataTrain.Y-alpha); u2=((G'*G+v2*eye(n+1,n+1))\G')*(DataTrain.Y-gamma); else Q1=eye(m,m)/c1+G*((G'*G+v1*eye(m+1,m+1))\G'); Q2=eye(m,m)/c2+G*((G'*G+v2*eye(m+1,m+1))\G'); p1=(eye(m,m)-G*((G'*G+v1*eye(m+1,m+1))\G'))*DataTrain.Y+eps1*e; p2=(-eye(m,m)+G*((G'*G+v2*eye(m+1,m+1))\G'))*DataTrain.Y+eps2*e; alpha=(Q1)\e;oldalpha=alpha+1;it1=0; gamma=(Q2)\e;oldgamma=gamma+1;it2=0; while it1tol1 z1= Q1*alpha+p1-beta1*alpha; pl=(abs(z1)+z1)/2-p1; oldalpha=alpha; alpha=Q1\pl; it1=it1+1; end; while it2tol2 z2= Q2*gamma+p2-beta2*gamma; pl=(abs(z2)+z2)/2-p2; oldgamma=gamma; gamma=Q2\pl; it2=it2+1; end; u1=(G'*G+v1*eye(m+1,m+1))\G'*(DataTrain.Y-alpha); u2=((G'*G+v2*eye(m+1,m+1))\G')*(DataTrain.Y-gamma); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Predict and output %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% e=ones(size(TestX,1),1); m = size(u1,1); if strcmp(kerfPara.type,'lin') Y1=TestX*u1(1:m-1)+u1(m)*e; Y2=TestX*u2(1:m-1)+u2(m)*e; else H=kernelfun(TestX,kerfPara,DataTrain.X); Y1=H*u1(1:m-1)+u1(m)*e; Y2=H*u2(1:m-1)+u2(m)*e; end DarwY.Y1 = Y1 - eps1; DarwY.Y2 = Y2 + eps1; PredictY=0.5*(Y1+Y2); end

ETSVR

Reference

Main Function

Contacts