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.ml.pca;
031
032import java.util.Collection;
033
034import org.openimaj.feature.DoubleFV;
035import org.openimaj.feature.FeatureVector;
036import org.openimaj.math.matrix.algorithm.pca.PrincipalComponentAnalysis;
037import org.openimaj.math.matrix.algorithm.pca.SvdPrincipalComponentAnalysis;
038
039import Jama.Matrix;
040
041/**
042 * Principal Components Analysis wrapper for {@link FeatureVector}s.
043 * 
044 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
045 */
046public class FeatureVectorPCA extends PrincipalComponentAnalysis {
047        PrincipalComponentAnalysis inner;
048
049        /**
050         * Default constructor, using an {@link SvdPrincipalComponentAnalysis}.
051         */
052        public FeatureVectorPCA() {
053                this.inner = new SvdPrincipalComponentAnalysis();
054        }
055
056        /**
057         * Construct with the given {@link PrincipalComponentAnalysis} object.
058         * 
059         * @param inner
060         *            PCA algorithm.
061         */
062        public FeatureVectorPCA(PrincipalComponentAnalysis inner) {
063                this.inner = inner;
064        }
065
066        /**
067         * Learn the PCA basis of the given feature vectors.
068         * 
069         * @param data
070         *            the feature vectors to apply PCA to.
071         */
072        public void learnBasis(FeatureVector[] data) {
073                final double[][] d = new double[data.length][];
074
075                for (int i = 0; i < data.length; i++) {
076                        d[i] = data[i].asDoubleVector();
077                }
078
079                learnBasis(d);
080        }
081
082        /**
083         * Learn the PCA basis of the given feature vectors.
084         * 
085         * @param data
086         *            the feature vectors to apply PCA to.
087         */
088        public void learnBasis(Collection<? extends FeatureVector> data) {
089                final double[][] d = new double[data.size()][];
090
091                int i = 0;
092                for (final FeatureVector fv : data) {
093                        d[i++] = fv.asDoubleVector();
094                }
095
096                learnBasis(d);
097        }
098
099        /**
100         * Project a vector by the basis. The vector is normalised by subtracting
101         * the mean and then multiplied by the basis.
102         * 
103         * @param vector
104         *            the vector to project
105         * @return projected vector
106         */
107        public DoubleFV project(FeatureVector vector) {
108                return new DoubleFV(project(vector.asDoubleVector()));
109        }
110
111        @Override
112        public void learnBasis(double[][] data) {
113                inner.learnBasis(data);
114                this.basis = inner.getBasis();
115                this.eigenvalues = inner.getEigenValues();
116                this.mean = inner.getMean();
117        }
118
119        @Override
120        protected void learnBasisNorm(Matrix norm) {
121                inner.learnBasis(norm);
122        }
123
124        /**
125         * Generate a new "observation" as a linear combination of the principal
126         * components (PC): mean + PC * scaling.
127         * 
128         * If the scaling vector is shorter than the number of components, it will
129         * be zero-padded. If it is longer, it will be truncated.
130         * 
131         * @param scalings
132         *            the weighting for each PC
133         * @return generated observation
134         */
135        public DoubleFV generate(DoubleFV scalings) {
136                return new DoubleFV(generate(scalings.values));
137        }
138}