|
- % Example of how to use the mklreg
- %
-
- clear all
- close all
- clc
- %-------------------creating and plotting data----------------
- n=100;
- bruit=0.3;
- freq=0.8;
- x=linspace(0,8,n)';
- xtest=linspace(0,8,n)';
- xapp=x;
- yapp=cos(exp(freq*x)) +randn(n,1)*bruit;
- ytest=cos(exp(freq*xtest));
-
-
- %----------------------Learning Parameters -------------------
- C = 100;
- verbose=1;
- options.algo='svmreg';
- options.seuildiffsigma=1e-3;
- options.seuildiffconstraint=0.01;
- options.seuildualitygap=0.01;
- options.goldensearch_deltmax=1e-2;
- options.numericalprecision=1e-8;
- options.stopvariation=0;
- options.stopKKT=0;
- options.stopdualitygap=1;
- options.firstbasevariable='first';
- options.nbitermax=500;
- options.seuil=0;
- options.seuilitermax=10;
- options.lambdareg = 1e-8;
- options.miniter=0;
- options.verbosesvm=0;
- options.svmreg_epsilon=0.01;
- options.efficientkernel=0;
- optionK.pow=0;
-
- kernelt={'gaussian' 'poly'};
- kerneloptionvect={[0.01:0.05:0.2 0.5 1 2 5 7 10 12 15 17 20] [1 2 3]};
- variablevec={'all' 'all'};
- dim=size(xapp,2);
-
- [kernel,kerneloptionvec,optionK.variablecell]=CreateKernelListWithVariable(variablevec,dim,kernelt,kerneloptionvect);
- [K]=mklbuildkernel(xapp,kernel,kerneloptionvec,[],[],optionK);
- [K,optionK.weightK]=WeightK(K);
-
-
-
- [beta,w,b,posw,story,obj] = mklsvm(K,yapp,C,options,verbose);
- kerneloption.matrix=mklbuildkernel(xtest,kernel,kerneloptionvec,xapp(posw,:),beta,optionK);
- ypred=svmval([],[],w,b,'numerical',kerneloption);
- plot(xtest,ytest,'b',xapp,yapp,'r',xapp,yapp,'r+',xtest,ypred,'g')
|