wolffd@0: function net = gpinit(net, tr_in, tr_targets, prior)
wolffd@0: %GPINIT	Initialise Gaussian Process model.
wolffd@0: %
wolffd@0: %	Description
wolffd@0: %	NET = GPINIT(NET, TRIN, TRTARGETS) takes a Gaussian Process data
wolffd@0: %	structure NET  together  with a matrix TRIN of training input vectors
wolffd@0: %	and a matrix TRTARGETS of  training target vectors, and stores them
wolffd@0: %	in NET. These datasets are required if the corresponding inverse
wolffd@0: %	covariance matrix is not supplied to GPFWD. This is important if the
wolffd@0: %	data structure is saved and then reloaded before calling GPFWD. Each
wolffd@0: %	row of TRIN corresponds to one input vector and each row of TRTARGETS
wolffd@0: %	corresponds to one target vector.
wolffd@0: %
wolffd@0: %	NET = GPINIT(NET, TRIN, TRTARGETS, PRIOR) additionally initialises
wolffd@0: %	the parameters in NET from the PRIOR data structure which contains
wolffd@0: %	the mean and variance of the Gaussian distribution which is sampled
wolffd@0: %	from.
wolffd@0: %
wolffd@0: %	See also
wolffd@0: %	GP, GPFWD
wolffd@0: %
wolffd@0: 
wolffd@0: %	Copyright (c) Ian T Nabney (1996-2001)
wolffd@0: 
wolffd@0: errstring = consist(net, 'gp', tr_in, tr_targets);
wolffd@0: if ~isempty(errstring);
wolffd@0:   error(errstring);
wolffd@0: end
wolffd@0: 
wolffd@0: if nargin >= 4 
wolffd@0:   % Initialise weights at random
wolffd@0:   if size(prior.pr_mean) == [1 1]
wolffd@0:     w = randn(1, net.nwts).*sqrt(prior.pr_var) + ...
wolffd@0:        repmat(prior.pr_mean, 1, net.nwts);
wolffd@0:   else
wolffd@0:     sig = sqrt(prior.index*prior.pr_var);
wolffd@0:     w = sig'.*randn(1, net.nwts) + (prior.index*prior.pr_mean)'; 
wolffd@0:   end
wolffd@0:   net = gpunpak(net, w);
wolffd@0: end
wolffd@0: 
wolffd@0: net.tr_in = tr_in;
wolffd@0: net.tr_targets = tr_targets;