berek_baygam.m

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

function [resultaat, gradient] = berek_evidence_gamma_gradR(gam)
% function [resultaat, gradient] = berek_evidence_gamma_gradR(gam)

gam = abs(gam)+100*eps; 
global sigm kern Xtr Ytr KK roK ALFA BETA EW ED EffPar vK;

LL= vK'*(Ytr-mean(Ytr));
N = size(Xtr,1); Neff = length(roK); 
som_log = sum(log(roK*gam+1)) - (N-1)*log(gam);

Ewgd = (LL.*1./(roK+1/gam))'*LL/2;
Ed = (LL.*1./(roK+1/gam).^2)'*LL/2/gam^2; Ew = Ewgd-gam*Ed;

resultaat = som_log + (N-1)*log( Ewgd );
gradient =  sum(roK./(roK*gam+1)) - (N-1)/(gam) + (N-1)/(Ew+gam*Ed)*Ed;

% Doe output naar de rest
ALFA = (N-1)/(2*(Ew+gam*Ed) );
BETA = gam*ALFA;
ED = Ed;
EW = Ew;