alphaprobs.m

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

function [pall,ff]=alphaprobs(X,y,alpha,pars,Xt);
%Determine the (tentative) probabilities produces by the current model
%maximum size of the kernel_matrix [maxsize,maxsize]
maxsize=3000;
N=size(X,1);
K=length(unique(y));c=K-1;
nr=ceil(N/maxsize);

ff=zeros(N,c);

for i=1:nr,
    if (i==nr),
        omegaSmall = kernel_matrix(X((i-1)*maxsize + 1 : end,:),pars.ker,pars.sig2,Xt);       
        ff((i-1)*maxsize+1:end,:)=omegaSmall*alpha;
    else
        omegaSmall = kernel_matrix(X((i-1)*maxsize + 1 : i*maxsize,:),pars.ker,pars.sig2,Xt);       
        ff((i-1)*maxsize+1:i*maxsize,:)=omegaSmall*alpha;
    end
end

denom=ones(size(ff,1),1)+sum(exp(ff),2);
pall=exp(ff)./repmat(denom,1,c);