Daniel@0: function [data, label] = gmmsamp(mix, n)
Daniel@0: %GMMSAMP Sample from a Gaussian mixture distribution.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %
Daniel@0: %	DATA = GSAMP(MIX, N) generates a sample of size N from a Gaussian
Daniel@0: %	mixture distribution defined by the MIX data structure. The matrix X
Daniel@0: %	has N rows in which each row represents a MIX.NIN-dimensional sample
Daniel@0: %	vector.
Daniel@0: %
Daniel@0: %	[DATA, LABEL] = GMMSAMP(MIX, N) also returns a column vector of
Daniel@0: %	classes (as an index 1..N) LABEL.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	GSAMP, GMM
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: % Check input arguments
Daniel@0: errstring = consist(mix, 'gmm');
Daniel@0: if ~isempty(errstring)
Daniel@0:   error(errstring);
Daniel@0: end
Daniel@0: if n < 1
Daniel@0:   error('Number of data points must be positive')
Daniel@0: end
Daniel@0: 
Daniel@0: % Determine number to sample from each component
Daniel@0: priors = rand(1, n);
Daniel@0: 
Daniel@0: % Pre-allocate data array
Daniel@0: data = zeros(n, mix.nin);
Daniel@0: if nargout > 1
Daniel@0:   label = zeros(n, 1);
Daniel@0: end
Daniel@0: cum_prior = 0;		% Cumulative sum of priors
Daniel@0: total_samples = 0;	% Cumulative sum of number of sampled points
Daniel@0: for j = 1:mix.ncentres
Daniel@0:   num_samples = sum(priors >= cum_prior & ...
Daniel@0:     priors < cum_prior + mix.priors(j));
Daniel@0:   % Form a full covariance matrix
Daniel@0:   switch mix.covar_type
Daniel@0:     case 'spherical'
Daniel@0:       covar = mix.covars(j) * eye(mix.nin);
Daniel@0:     case 'diag'
Daniel@0:       covar = diag(mix.covars(j, :));
Daniel@0:     case 'full'
Daniel@0:       covar = mix.covars(:, :, j);
Daniel@0:     case 'ppca'
Daniel@0:       covar = mix.covars(j) * eye(mix.nin) + ...
Daniel@0:         mix.U(:, :, j)* ...
Daniel@0:         (diag(mix.lambda(j, :))-(mix.covars(j)*eye(mix.ppca_dim)))* ...
Daniel@0:         (mix.U(:, :, j)');
Daniel@0:     otherwise
Daniel@0:       error(['Unknown covariance type ', mix.covar_type]);
Daniel@0:   end
Daniel@0:   data(total_samples+1:total_samples+num_samples, :) = ...
Daniel@0:     gsamp(mix.centres(j, :), covar, num_samples);
Daniel@0:   if nargout > 1
Daniel@0:     label(total_samples+1:total_samples+num_samples) = j;
Daniel@0:   end
Daniel@0:   cum_prior = cum_prior + mix.priors(j);
Daniel@0:   total_samples = total_samples + num_samples;
Daniel@0: end
Daniel@0: