wolffd@0: function [mu, Sigma, weights] = mixgauss_init(M, data, cov_type, method)
wolffd@0: % MIXGAUSS_INIT Initial parameter estimates for a mixture of Gaussians
wolffd@0: % function [mu, Sigma, weights] = mixgauss_init(M, data, cov_type. method)
wolffd@0: %
wolffd@0: % INPUTS:
wolffd@0: % data(:,t) is the t'th example
wolffd@0: % M = num. mixture components
wolffd@0: % cov_type = 'full', 'diag' or 'spherical'
wolffd@0: % method = 'rnd' (choose centers randomly from data) or 'kmeans' (needs netlab)
wolffd@0: %
wolffd@0: % OUTPUTS:
wolffd@0: % mu(:,k) 
wolffd@0: % Sigma(:,:,k) 
wolffd@0: % weights(k)
wolffd@0: 
wolffd@0: if nargin < 4, method = 'kmeans'; end
wolffd@0: 
wolffd@0: [d T] = size(data);
wolffd@0: data = reshape(data, d, T); % in case it is data(:, t, sequence_num)
wolffd@0: 
wolffd@0: switch method
wolffd@0:  case 'rnd', 
wolffd@0:   C = cov(data');
wolffd@0:   Sigma = repmat(diag(diag(C))*0.5, [1 1 M]);
wolffd@0:   % Initialize each mean to a random data point
wolffd@0:   indices = randperm(T);
wolffd@0:   mu = data(:,indices(1:M));
wolffd@0:   weights = normalise(ones(M,1));
wolffd@0:  case 'kmeans',
wolffd@0:   mix = gmm(d, M, cov_type);
wolffd@0:   options = foptions;
wolffd@0:   max_iter = 5;
wolffd@0:   options(1) = -1; % be quiet!
wolffd@0:   options(14) = max_iter;
wolffd@0:   mix = gmminit(mix, data', options);
wolffd@0:   mu = reshape(mix.centres', [d M]);
wolffd@0:   weights = mix.priors(:);
wolffd@0:   for m=1:M
wolffd@0:     switch cov_type
wolffd@0:      case 'diag',
wolffd@0:       Sigma(:,:,m) = diag(mix.covars(m,:));
wolffd@0:      case 'full',
wolffd@0:       Sigma(:,:,m) = mix.covars(:,:,m);
wolffd@0:      case 'spherical',
wolffd@0:       Sigma(:,:,m) = mix.covars(m) * eye(d);
wolffd@0:     end
wolffd@0:   end
wolffd@0: end
wolffd@0: