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.clustering.spectral;
031
032import org.openimaj.data.DataSource;
033import org.openimaj.knn.DoubleNearestNeighbours;
034import org.openimaj.knn.DoubleNearestNeighboursExact;
035import org.openimaj.ml.clustering.SpatialClusterer;
036import org.openimaj.ml.clustering.SpatialClusters;
037import org.openimaj.ml.clustering.dbscan.DoubleNNDBSCAN;
038import org.openimaj.util.function.Function;
039import org.openimaj.util.pair.IndependentPair;
040
041import ch.akuhn.matrix.eigenvalues.Eigenvalues;
042
043/**
044 * @author Sina Samangooei (ss@ecs.soton.ac.uk)
045 * @param <DATATYPE>
046 *
047 */
048public class SpectralClusteringConf<DATATYPE>{
049        /**
050         * A function which can represent itself as a string
051         *
052         * @param <DATATYPE>
053         * @author Sina Samangooei (ss@ecs.soton.ac.uk)
054         */
055        public static interface ClustererProvider<DATATYPE> extends Function<IndependentPair<double[], double[][]>, SpatialClusterer<? extends SpatialClusters<DATATYPE>,DATATYPE>>{
056                public String toString();
057        }
058        protected static class DefaultClustererFunction<DATATYPE> implements ClustererProvider<DATATYPE>{
059
060
061                private SpatialClusterer<? extends SpatialClusters<DATATYPE>, DATATYPE> internal;
062
063                public DefaultClustererFunction( SpatialClusterer<? extends SpatialClusters<DATATYPE>,DATATYPE> internal) {
064                        this.internal = internal;
065                }
066
067                @Override
068                public SpatialClusterer<? extends SpatialClusters<DATATYPE>, DATATYPE> apply(IndependentPair<double[], double[][]> in) {
069                        return internal;
070                }
071                
072                @Override
073                public String toString() {
074                        return internal.toString();
075                }
076                
077        }
078        
079        /**
080         * The internal clusterer
081         */
082        
083        ClustererProvider<DATATYPE> internal;
084
085        /**
086         * The graph laplacian creator
087         */
088        public GraphLaplacian laplacian;
089
090        /**
091         * The method used to select the number of eigen vectors from the lower valued eigenvalues
092         */
093        public EigenChooser eigenChooser;
094
095        public int skipEigenVectors = 0;
096
097        public boolean eigenValueScale = false;
098
099        /**
100         * @param internal the internal clusterer
101         * @param eigK the value handed to {@link HardCodedEigenChooser}
102         *
103         */
104        public SpectralClusteringConf(SpatialClusterer<? extends SpatialClusters<DATATYPE>,DATATYPE> internal, int eigK) {
105                this.internal = new DefaultClustererFunction<DATATYPE>(internal);
106                this.laplacian = new GraphLaplacian.Normalised();
107                this.eigenChooser = new HardCodedEigenChooser(eigK);
108
109        }
110
111        /**
112         * The underlying {@link EigenChooser} is set to an {@link ChangeDetectingEigenChooser} which
113         * looks for a 100x gap between eigen vectors to select number of clusters. It also insists upon
114         * a maximum of 0.1 * number of data items (so 10 items per cluster)
115         *
116         * @param internal the internal clusterer
117         *
118         */
119        public SpectralClusteringConf(SpatialClusterer<? extends SpatialClusters<DATATYPE>,DATATYPE> internal) {
120                this.internal = new DefaultClustererFunction<DATATYPE>(internal);
121                this.laplacian = new GraphLaplacian.Normalised();
122                this.eigenChooser = new ChangeDetectingEigenChooser(100,0.1);
123
124        }
125        
126        /**
127         * @param internal an internal clusterer 
128         * @param lap the laplacian
129         * @param top the top eigen vectors
130         */
131        public SpectralClusteringConf(SpatialClusterer<? extends SpatialClusters<DATATYPE>, DATATYPE> internal, GraphLaplacian lap, int top) {
132                this.internal = new DefaultClustererFunction<DATATYPE>(internal);
133                this.laplacian = lap;
134                this.eigenChooser = new HardCodedEigenChooser(top);
135        }
136
137        /**
138         * The underlying {@link EigenChooser} is set to an {@link ChangeDetectingEigenChooser} which
139         * looks for a 100x gap between eigen vectors to select number of clusters. It also insists upon
140         * a maximum of 0.1 * number of data items (so 10 items per cluster)
141         *
142         * @param internal the internal clusterer
143         * @param laplacian the graph laplacian
144         *
145         */
146        public SpectralClusteringConf(SpatialClusterer<? extends SpatialClusters<DATATYPE>,DATATYPE> internal, GraphLaplacian laplacian) {
147                this.internal = new DefaultClustererFunction<DATATYPE>(internal);
148                this.laplacian = laplacian;
149                this.eigenChooser = new ChangeDetectingEigenChooser(100,0.1);
150
151        }
152        
153        /**
154         * The underlying {@link EigenChooser} is set to an {@link ChangeDetectingEigenChooser} which
155         * looks for a 100x gap between eigen vectors to select number of clusters. It also insists upon
156         * a maximum of 0.1 * number of data items (so 10 items per cluster)
157         * @param internal the internal clusterer
158         *
159         */
160        public SpectralClusteringConf(ClustererProvider<DATATYPE> internal) {
161                this.internal = internal;
162                this.laplacian = new GraphLaplacian.Normalised();
163                this.eigenChooser = new ChangeDetectingEigenChooser(100,0.1);
164
165        }
166
167        
168}