001/**
002 * Copyright (c) 2011, The University of Southampton and the individual contributors.
003 * All rights reserved.
004 *
005 * Redistribution and use in source and binary forms, with or without modification,
006 * are permitted provided that the following conditions are met:
007 *
008 *   *  Redistributions of source code must retain the above copyright notice,
009 *      this list of conditions and the following disclaimer.
010 *
011 *   *  Redistributions in binary form must reproduce the above copyright notice,
012 *      this list of conditions and the following disclaimer in the documentation
013 *      and/or other materials provided with the distribution.
014 *
015 *   *  Neither the name of the University of Southampton nor the names of its
016 *      contributors may be used to endorse or promote products derived from this
017 *      software without specific prior written permission.
018 *
019 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
020 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
021 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
022 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
023 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
024 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
025 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
026 * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
027 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
028 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
029 */
030package org.openimaj.math.matrix.algorithm.pca;
031
032import java.util.Arrays;
033
034import no.uib.cipr.matrix.NotConvergedException;
035import Jama.Matrix;
036
037
038/**
039 * Compute the PCA using an SVD without actually constructing
040 * the covariance matrix. This class performs a full SVD extracting all
041 * singular values and vectors. If you know apriori how many principle
042 * components (or have an upper bound on the number), then use a
043 * {@link ThinSvdPrincipalComponentAnalysis} instead. 
044 * 
045 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
046 */
047public class SvdPrincipalComponentAnalysis extends PrincipalComponentAnalysis {
048        int ndims;
049        
050        /**
051         * Construct a {@link SvdPrincipalComponentAnalysis} that
052         * will extract all the eigenvectors.
053         */
054        public SvdPrincipalComponentAnalysis() {
055                this(-1);
056        }
057        
058        /**
059         * Construct a {@link SvdPrincipalComponentAnalysis} that
060         * will extract the n best eigenvectors.
061         * @param ndims the number of eigenvectors to select.
062         */
063        public SvdPrincipalComponentAnalysis(int ndims) {
064                this.ndims = ndims;
065        }
066        
067        @Override
068        public void learnBasisNorm(Matrix norm) {
069                try {
070                        no.uib.cipr.matrix.DenseMatrix mjtA = new no.uib.cipr.matrix.DenseMatrix(norm.getArray());
071                        no.uib.cipr.matrix.SVD svd = no.uib.cipr.matrix.SVD.factorize(mjtA);
072                        
073                        no.uib.cipr.matrix.DenseMatrix output = svd.getVt();
074
075                        int dims = ndims < 0 ? svd.getS().length : ndims;
076
077                        basis = new Matrix(output.numColumns(), dims);
078                        eigenvalues = Arrays.copyOf(svd.getS(), dims);
079                        
080                        double normEig = 1.0 / (norm.getRowDimension() - 1);
081                        for (int i=0; i<eigenvalues.length; i++) 
082                                eigenvalues[i] = eigenvalues[i] * eigenvalues[i] * normEig;
083                        
084                        double[][] basisData = basis.getArray();
085                        for (int j=0; j<output.numColumns(); j++)
086                                for (int i=0; i<dims; i++)
087                                        basisData[j][i] = output.get(i, j);
088                        
089                } catch (NotConvergedException e) {
090                        throw new RuntimeException(e);
091                }
092        }
093}