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.experiment.validation;
031
032import java.util.Map;
033
034import org.openimaj.data.RandomData;
035import org.openimaj.data.dataset.GroupedDataset;
036import org.openimaj.data.dataset.ListBackedDataset;
037import org.openimaj.data.dataset.ListDataset;
038import org.openimaj.data.dataset.MapBackedDataset;
039import org.openimaj.util.pair.IndependentPair;
040
041/**
042 * Hold-Out validation for grouped data that selects a percentage of the
043 * original data to use for training, and the remainder to use for validation.
044 * No attempt is made to ensure that the distribution of relative group sizes is
045 * constant. If this is important, use a
046 * {@link StratifiedGroupedRandomisedPercentageHoldOut} instead.
047 * 
048 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
049 * 
050 * @param <KEY>
051 *            Type of groups
052 * @param <INSTANCE>
053 *            Type of the instances in the dataset
054 */
055public class GroupedRandomisedPercentageHoldOut<KEY, INSTANCE>
056                extends
057                DefaultValidationData<GroupedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>>
058{
059        /**
060         * Construct with the given dataset and percentage of training data (0..1).
061         * 
062         * @param percentageTraining
063         *            percentage of the dataset to use for training
064         * @param dataset
065         *            the dataset
066         */
067        public GroupedRandomisedPercentageHoldOut(double percentageTraining,
068                        GroupedDataset<KEY, ListDataset<INSTANCE>, INSTANCE> dataset)
069        {
070                if (percentageTraining < 0 || percentageTraining > 1)
071                        throw new IllegalArgumentException(
072                                        "percentage of training instances must be between 0 and 1");
073
074                final int size = dataset.numInstances();
075                final int[] indices = RandomData.getUniqueRandomInts(size, 0, size);
076                final int nTrain = (int) (percentageTraining * size);
077
078                this.training = new MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>();
079                final Map<KEY, ListDataset<INSTANCE>> trainMap = ((MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>) training)
080                                .getMap();
081                for (int i = 0; i < nTrain; i++) {
082                        final IndependentPair<KEY, INSTANCE> p = select(indices[i], dataset);
083
084                        ListBackedDataset<INSTANCE> lmd = (ListBackedDataset<INSTANCE>) trainMap
085                                        .get(p.firstObject());
086                        if (lmd == null)
087                                trainMap.put(p.firstObject(),
088                                                lmd = new ListBackedDataset<INSTANCE>());
089                        lmd.add(p.getSecondObject());
090                }
091
092                this.validation = new MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>();
093                final Map<KEY, ListDataset<INSTANCE>> validMap = ((MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>) validation)
094                                .getMap();
095                for (int i = nTrain; i < size; i++) {
096                        final IndependentPair<KEY, INSTANCE> p = select(indices[i], dataset);
097
098                        ListBackedDataset<INSTANCE> lmd = (ListBackedDataset<INSTANCE>) validMap
099                                        .get(p.firstObject());
100                        if (lmd == null)
101                                validMap.put(p.firstObject(),
102                                                lmd = new ListBackedDataset<INSTANCE>());
103                        lmd.add(p.getSecondObject());
104                }
105        }
106
107        private IndependentPair<KEY, INSTANCE> select(int idx,
108                        GroupedDataset<KEY, ListDataset<INSTANCE>, INSTANCE> dataset)
109        {
110                for (final KEY k : dataset.getGroups()) {
111                        final ListDataset<INSTANCE> instances = dataset.getInstances(k);
112                        final int sz = instances.size();
113
114                        if (idx < sz) {
115                                return new IndependentPair<KEY, INSTANCE>(k,
116                                                instances.getInstance(idx));
117                        }
118                        idx -= sz;
119                }
120
121                return null;
122        }
123}