predictklr.m

The goal of SPID is to provide the user with tools capable to simulate, preprocess, process and clas

function [lhte,AbsAcct,ysimte]=predictKLR(X,y,alpha,Xtest,ytest,pars)

X=[ones(size(X,1),1) X];
Xtest=[ones(size(Xtest,1),1) Xtest];
K=length(unique(y));c=K-1;
[N,d]=size(X);

alphamat=reshape(alpha,c,N)'; % matrix notation
p=alphaprobs(X,y,alphamat,pars,X);
pt=alphaprobs(Xtest,ytest,alphamat,pars,X);

%compute likelihood
lhtr=lh(y,p,pars);
lhte=lh(ytest,pt,pars);

p=[p ones(size(p,1),1)-sum(p,2)];
pt=[pt ones(size(pt,1),1)-sum(pt,2)];    

if pars.prior==1
    atrain=leesAantal(y);
    priors=diff(atrain)/sum(diff(atrain));
    p=repmat(priors,size(p,1),1).*p;
    pt=repmat(priors,size(pt,1),1).*pt;
end

[m,ysimtr]=max(p');
[mt,ysimte]=max(pt');
AbsAcc=100*sum(ysimtr' == y)/length(y);
AbsAcct=100*sum(ysimte' == ytest)/length(ytest);

if (pars.info==1)
    disp(['TRAIN: Accuracy ' num2str(AbsAcc) ' percentage, likelihood (norm):' num2str(lhtr)]);
    disp(['TEST : Accuracy ' num2str(AbsAcct) ' percentage, likelihood (norm):' num2str(lhte)]);
end