|
wolffd@0
|
1 function mixgauss = mixgauss_classifier_train(trainFeatures, trainLabels, nc, varargin)
|
|
wolffd@0
|
2 % function mixgauss = mixgauss_classifier_train(trainFeatures, trainLabels, nclusters, varargin)
|
|
wolffd@0
|
3 % trainFeatures(:,i) for i'th example
|
|
wolffd@0
|
4 % trainLabels should be 0,1
|
|
wolffd@0
|
5 % To evaluate performance on a tets set, use
|
|
wolffd@0
|
6 % mixgauss = mixgauss_classifier_train(trainFeatures, trainLabels, nc, 'testFeatures', tf, 'testLabels', tl)
|
|
wolffd@0
|
7
|
|
wolffd@0
|
8 [testFeatures, testLabels, max_iter, thresh, cov_type, mu, Sigma, priorC, method, ...
|
|
wolffd@0
|
9 cov_prior, verbose, prune_thresh] = process_options(...
|
|
wolffd@0
|
10 varargin, 'testFeatures', [], 'testLabels', [], ...
|
|
wolffd@0
|
11 'max_iter', 10, 'thresh', 0.01, 'cov_type', 'diag', ...
|
|
wolffd@0
|
12 'mu', [], 'Sigma', [], 'priorC', [], 'method', 'kmeans', ...
|
|
wolffd@0
|
13 'cov_prior', [], 'verbose', 0, 'prune_thresh', 0);
|
|
wolffd@0
|
14
|
|
wolffd@0
|
15 Nclasses = 2; % max([trainLabels testLabels]) + 1;
|
|
wolffd@0
|
16
|
|
wolffd@0
|
17 pos = find(trainLabels == 1);
|
|
wolffd@0
|
18 neg = find(trainLabels == 0);
|
|
wolffd@0
|
19
|
|
wolffd@0
|
20 if verbose, fprintf('fitting pos\n'); end
|
|
wolffd@0
|
21 [mixgauss.pos.mu, mixgauss.pos.Sigma, mixgauss.pos.prior] = ...
|
|
wolffd@0
|
22 mixgauss_em(trainFeatures(:, pos), nc, varargin{:});
|
|
wolffd@0
|
23
|
|
wolffd@0
|
24 if verbose, fprintf('fitting neg\n'); end
|
|
wolffd@0
|
25 [mixgauss.neg.mu, mixgauss.neg.Sigma, mixgauss.neg.prior] = ...
|
|
wolffd@0
|
26 mixgauss_em(trainFeatures(:, neg), nc, varargin{:});
|
|
wolffd@0
|
27
|
|
wolffd@0
|
28
|
|
wolffd@0
|
29 if ~isempty(priorC)
|
|
wolffd@0
|
30 mixgauss.priorC = priorC;
|
|
wolffd@0
|
31 else
|
|
wolffd@0
|
32 mixgauss.priorC = normalize([length(pos) length(neg)]);
|
|
wolffd@0
|
33 end
|
