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.dataset.GroupedDataset;
035import org.openimaj.data.dataset.ListDataset;
036import org.openimaj.data.dataset.MapBackedDataset;
037
038/**
039 * Stratified Hold-Out validation for grouped data that selects a percentage of
040 * the original data to use for training, and the remainder to use for
041 * validation. The splitting of data is performed per group to ensure that the
042 * relative group sizes remain (approximately) constant.
043 * 
044 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
045 * 
046 * @param <KEY>
047 *            Type of groups
048 * @param <INSTANCE>
049 *            Type of the instances in the dataset
050 */
051public class StratifiedGroupedRandomisedPercentageHoldOut<KEY, INSTANCE>
052                extends
053                DefaultValidationData<GroupedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>> {
054
055        /**
056         * Construct with the given dataset and percentage of training data (0..1).
057         * 
058         * @param percentageTraining
059         *            percentage of the dataset to use for training
060         * @param dataset
061         *            the dataset
062         */
063        public StratifiedGroupedRandomisedPercentageHoldOut(
064                        double percentageTraining,
065                        GroupedDataset<KEY, ListDataset<INSTANCE>, INSTANCE> dataset) {
066                if (percentageTraining < 0 || percentageTraining > 1)
067                        throw new IllegalArgumentException(
068                                        "percentage of training instances must be between 0 and 1");
069
070                this.training = new MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>();
071                this.validation = new MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>();
072
073                Map<KEY, ListDataset<INSTANCE>> trainMap = ((MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>) training)
074                                .getMap();
075                Map<KEY, ListDataset<INSTANCE>> validMap = ((MapBackedDataset<KEY, ListDataset<INSTANCE>, INSTANCE>) validation)
076                                .getMap();
077
078                for (KEY key : dataset.getGroups()) {
079                        RandomisedPercentageHoldOut<INSTANCE> ho = new RandomisedPercentageHoldOut<INSTANCE>(
080                                        percentageTraining, dataset.getInstances(key));
081
082                        trainMap.put(key, ho.training);
083                        validMap.put(key, ho.validation);
084                }
085        }
086}