camir-aes2014: toolboxes/FullBNT-1.0.7/KPMtools/pca

first hg version after svn

rev	line source
wolffd@0	1 function [pc_vec]=pca_kpm(features,N, method);
wolffd@0	2 % PCA_KPM Compute top N principal components using eigs or svd.
wolffd@0	3 % [pc_vec]=pca_kpm(features,N)
wolffd@0	4 %
wolffd@0	5 % features(:,i) is the i'th example - each COLUMN is an observation
wolffd@0	6 % pc_vec(:,j) is the j'th basis function onto which you should project the data
wolffd@0	7 % using pc_vec' * features
wolffd@0	8
wolffd@0	9 [d ncases] = size(features);
wolffd@0	10 fm=features-repmat(mean(features,2), 1, ncases);
wolffd@0	11
wolffd@0	12
wolffd@0	13 if method==1 % dd < dncases
wolffd@0	14 fprintf('pca_kpm eigs\n');
wolffd@0	15 options.disp = 0;
wolffd@0	16 C = cov(fm'); % d x d matrix
wolffd@0	17 [pc_vec, evals] = eigs(C, N, 'LM', options);
wolffd@0	18 else
wolffd@0	19 % [U,D,V] = SVD(fm), U(:,i)=evec of fm fm', V(:,i) = evec of fm' fm
wolffd@0	20 fprintf('pca_kpm svds\n');
wolffd@0	21 [U,D,V] = svds(fm', N);
wolffd@0	22 pc_vec = V;
wolffd@0	23 end
wolffd@0	24
wolffd@0	25 if 0
wolffd@0	26 X = randn(5,3);
wolffd@0	27 X = X-repmat(mean(X),5,1);
wolffd@0	28 C = X'*X;
wolffd@0	29 C2=cov(X)
wolffd@0	30 [U,D,V]=svd(X);
wolffd@0	31 [V2,D2]=eig(X)
wolffd@0	32 end

Mercurial > hg > camir-aes2014