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;
031
032import org.openimaj.data.dataset.GroupedDataset;
033import org.openimaj.data.dataset.ListDataset;
034import org.openimaj.ml.training.IncrementalTrainer;
035
036/**
037 * An {@link Annotator} that can be trained/updated incrementally.
038 * 
039 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
040 * 
041 * @param <OBJECT>
042 *            Type of object
043 * @param <ANNOTATION>
044 *            Type of annotation
045 */
046public abstract class IncrementalAnnotator<OBJECT, ANNOTATION>
047                extends
048                AbstractAnnotator<OBJECT, ANNOTATION>
049                implements
050                IncrementalTrainer<Annotated<OBJECT, ANNOTATION>>
051{
052        protected IncrementalAnnotator() {
053        }
054
055        /**
056         * Train the annotator with the given data. The default implementation of
057         * this method just calls {@link #train(Object)} on each data item.
058         * Subclasses may override to do something more intelligent if necessary.
059         * 
060         * @param data
061         *            the training data
062         */
063        @Override
064        public void train(Iterable<? extends Annotated<OBJECT, ANNOTATION>> data) {
065                for (final Annotated<OBJECT, ANNOTATION> d : data)
066                        train(d);
067        }
068
069        /**
070         * Train the annotator with the given grouped dataset. This method assumes
071         * that each object only appears in a <b>single</b> group of the dataset
072         * (i.e. a multi-class problem). Each group corresponds to the one single
073         * annotation assigned to each object.
074         * <p>
075         * If your dataset contains multiple labels for each object (through an
076         * object appearing in multiple groups) you should use
077         * {@link #train(GroupedDataset)}.
078         * 
079         * @param dataset
080         *            the dataset to train on
081         */
082        public void trainMultiClass(GroupedDataset<ANNOTATION, ? extends ListDataset<OBJECT>, OBJECT> dataset) {
083                for (final ANNOTATION grp : dataset.getGroups()) {
084                        for (final OBJECT inst : dataset.getInstances(grp)) {
085                                train(new AnnotatedObject<OBJECT, ANNOTATION>(inst, grp));
086                        }
087                }
088        }
089
090        /**
091         * Train the annotator with the given grouped dataset. This method assumes
092         * that each object can appear in multiple groups of the dataset (i.e. a
093         * multi-label problem). Internally, the dataset is converted to a list
094         * containing exactly one reference to each object in the dataset with
095         * (potentially) multiple annotations.
096         * <p>
097         * If the dataset is actually multi-class (i.e. each object belongs to only
098         * a single group), then calling this method is equivalent to calling
099         * {@link #trainMultiClass(GroupedDataset)}, but is less efficient as the
100         * dataset has to be converted into a list.
101         * <p>
102         * Some annotator implementations do not care whether the data is
103         * multi-class or multi-label, and might choose to override this method to
104         * just call {@link #trainMultiClass(GroupedDataset)} instead.
105         * 
106         * @param dataset
107         *            the dataset to train on
108         */
109        public void train(GroupedDataset<ANNOTATION, ? extends ListDataset<OBJECT>, OBJECT> dataset) {
110                for (final AnnotatedObject<OBJECT, ANNOTATION> ao : AnnotatedObject.createList(dataset)) {
111                        train(ao);
112                }
113        }
114}