wolffd@0: function net = rbfsetbf(net, options, x)
wolffd@0: %RBFSETBF Set basis functions of RBF from data.
wolffd@0: %
wolffd@0: %	Description
wolffd@0: %	NET = RBFSETBF(NET, OPTIONS, X) sets the basis functions of the RBF
wolffd@0: %	network NET so that they model the unconditional density of the
wolffd@0: %	dataset X.  This is done by training a GMM with spherical covariances
wolffd@0: %	using GMMEM.  The OPTIONS vector is passed to GMMEM. The widths of
wolffd@0: %	the functions are set by a call to RBFSETFW.
wolffd@0: %
wolffd@0: %	See also
wolffd@0: %	RBFTRAIN, RBFSETFW, GMMEM
wolffd@0: %
wolffd@0: 
wolffd@0: %	Copyright (c) Ian T Nabney (1996-2001)
wolffd@0: 
wolffd@0: errstring = consist(net, 'rbf', x);
wolffd@0: if ~isempty(errstring)
wolffd@0:   error(errstring);
wolffd@0: end
wolffd@0: 
wolffd@0: % Create a spherical Gaussian mixture model
wolffd@0: mix = gmm(net.nin, net.nhidden, 'spherical');
wolffd@0: 
wolffd@0: % Initialise the parameters from the input data
wolffd@0: % Just use a small number of k means iterations
wolffd@0: kmoptions = zeros(1, 18);
wolffd@0: kmoptions(1) = -1;	% Turn off warnings
wolffd@0: kmoptions(14) = 5;  % Just 5 iterations to get centres roughly right
wolffd@0: mix = gmminit(mix, x, kmoptions);
wolffd@0: 
wolffd@0: % Train mixture model using EM algorithm
wolffd@0: [mix, options] = gmmem(mix, x, options);
wolffd@0: 
wolffd@0: % Now set the centres of the RBF from the centres of the mixture model
wolffd@0: net.c = mix.centres;
wolffd@0: 
wolffd@0: % options(7) gives scale of function widths
wolffd@0: net = rbfsetfw(net, options(7));