Daniel@0: function [e, edata, eprior] = gperr(net, x, t)
Daniel@0: %GPERR	Evaluate error function for Gaussian Process.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %	E = GPERR(NET, X, T) takes a Gaussian Process data structure NET
Daniel@0: %	together  with a matrix X of input vectors and a matrix T of target
Daniel@0: %	vectors, and evaluates the error function E. Each row of X
Daniel@0: %	corresponds to one input vector and each row of T corresponds to one
Daniel@0: %	target vector.
Daniel@0: %
Daniel@0: %	[E, EDATA, EPRIOR] = GPERR(NET, X, T) additionally returns the data
Daniel@0: %	and hyperprior components of the error, assuming a Gaussian prior on
Daniel@0: %	the weights with mean and variance parameters PRMEAN and PRVARIANCE
Daniel@0: %	taken from the network data structure NET.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	GP, GPCOVAR, GPFWD, GPGRAD
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: errstring = consist(net, 'gp', x, t);
Daniel@0: if ~isempty(errstring);
Daniel@0:   error(errstring);
Daniel@0: end
Daniel@0: 
Daniel@0: cn = gpcovar(net, x);
Daniel@0: 
Daniel@0: edata = 0.5*(sum(log(eig(cn, 'nobalance'))) + t'*inv(cn)*t);
Daniel@0: 
Daniel@0: % Evaluate the hyperprior contribution to the error.
Daniel@0: % The hyperprior is Gaussian with mean pr_mean and variance
Daniel@0: % pr_variance
Daniel@0: if isfield(net, 'pr_mean')
Daniel@0:   w = gppak(net);
Daniel@0:   m = repmat(net.pr_mean, size(w));
Daniel@0:   if size(net.pr_mean) == [1 1]
Daniel@0:     eprior = 0.5*((w-m)*(w-m)');
Daniel@0:     e2 = eprior/net.pr_var;
Daniel@0:   else
Daniel@0:     wpr = repmat(w, size(net.pr_mean, 1), 1)';
Daniel@0:     eprior = 0.5*(((wpr - m').^2).*net.index);
Daniel@0:     e2 = (sum(eprior, 1))*(1./net.pr_var);
Daniel@0:   end
Daniel@0: else
Daniel@0:   e2 = 0;
Daniel@0:   eprior = 0;
Daniel@0: end
Daniel@0: 
Daniel@0: e = edata + e2;
Daniel@0: