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.annotation.basic.util;
031
032import gnu.trove.map.hash.TIntIntHashMap;
033
034import java.util.Collection;
035import java.util.List;
036
037import org.openimaj.ml.annotation.Annotated;
038
039import cern.jet.random.Empirical;
040import cern.jet.random.EmpiricalWalker;
041import cern.jet.random.engine.MersenneTwister;
042
043/**
044 * Choose the number of annotations based on the numbers of annotations
045 * of each training example. Internally this {@link NumAnnotationsChooser}
046 * constructs the distribution of annotation lengths from the training
047 * data and then picks lengths randomly based on the distribution.
048 * 
049 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
050 */
051public class PriorChooser implements NumAnnotationsChooser {
052        private EmpiricalWalker numAnnotations;
053
054        @Override
055        public <O, A> void train(List<? extends Annotated<O, A>> data) {
056                TIntIntHashMap nAnnotationCounts = new TIntIntHashMap();
057                int maxVal = 0;
058                
059                for (Annotated<O, A> sample : data) {
060                        Collection<A> annos = sample.getAnnotations();
061
062                        nAnnotationCounts.adjustOrPutValue(annos.size(), 1, 1);
063                        
064                        if (annos.size()>maxVal) maxVal = annos.size();
065                }
066                
067                //build distribution and rng for the number of annotations
068                double [] distr = new double[maxVal+1];
069                for (int i=0; i<=maxVal; i++) 
070                        distr[i] = nAnnotationCounts.get(i);
071                numAnnotations = new EmpiricalWalker(distr, Empirical.NO_INTERPOLATION, new MersenneTwister());
072        }
073
074        @Override
075        public int numAnnotations() {
076                return numAnnotations.nextInt();
077        }
078}