Daniel@0: function mixgauss = mixgauss_classifier_train(trainFeatures, trainLabels, nc, varargin)
Daniel@0: % function mixgauss = mixgauss_classifier_train(trainFeatures, trainLabels, nclusters, varargin)
Daniel@0: % trainFeatures(:,i) for i'th example
Daniel@0: % trainLabels should be 0,1
Daniel@0: % To evaluate performance on a tets set, use
Daniel@0: % mixgauss = mixgauss_classifier_train(trainFeatures, trainLabels, nc, 'testFeatures', tf, 'testLabels', tl)
Daniel@0: 
Daniel@0: [testFeatures, testLabels, max_iter, thresh, cov_type, mu, Sigma, priorC, method, ...
Daniel@0:  cov_prior, verbose, prune_thresh] = process_options(...
Daniel@0:     varargin, 'testFeatures', [], 'testLabels', [], ...
Daniel@0:      'max_iter', 10, 'thresh', 0.01, 'cov_type', 'diag', ...
Daniel@0:     'mu', [], 'Sigma', [], 'priorC', [], 'method', 'kmeans', ...
Daniel@0:     'cov_prior', [], 'verbose', 0, 'prune_thresh', 0);
Daniel@0: 
Daniel@0: Nclasses = 2; % max([trainLabels testLabels]) + 1;
Daniel@0: 
Daniel@0: pos = find(trainLabels == 1);
Daniel@0: neg = find(trainLabels == 0);
Daniel@0: 
Daniel@0: if verbose, fprintf('fitting pos\n'); end
Daniel@0: [mixgauss.pos.mu, mixgauss.pos.Sigma, mixgauss.pos.prior] = ...
Daniel@0:     mixgauss_em(trainFeatures(:, pos), nc, varargin{:});
Daniel@0: 
Daniel@0: if verbose, fprintf('fitting neg\n'); end
Daniel@0: [mixgauss.neg.mu, mixgauss.neg.Sigma, mixgauss.neg.prior] = ...
Daniel@0:     mixgauss_em(trainFeatures(:, neg), nc, varargin{:});
Daniel@0: 
Daniel@0: 
Daniel@0: if ~isempty(priorC)
Daniel@0:   mixgauss.priorC = priorC;
Daniel@0: else
Daniel@0:   mixgauss.priorC = normalize([length(pos) length(neg)]);
Daniel@0: end