001 /*
002 * Java Genetic Algorithm Library (jenetics-7.1.1).
003 * Copyright (c) 2007-2022 Franz Wilhelmstötter
004 *
005 * Licensed under the Apache License, Version 2.0 (the "License");
006 * you may not use this file except in compliance with the License.
007 * You may obtain a copy of the License at
008 *
009 * http://www.apache.org/licenses/LICENSE-2.0
010 *
011 * Unless required by applicable law or agreed to in writing, software
012 * distributed under the License is distributed on an "AS IS" BASIS,
013 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
014 * See the License for the specific language governing permissions and
015 * limitations under the License.
016 *
017 * Author:
018 * Franz Wilhelmstötter (franz.wilhelmstoetter@gmail.com)
019 */
020 package io.jenetics.ext.moea;
021
022 import static java.lang.Math.min;
023 import static java.util.Objects.requireNonNull;
024
025 import java.util.ArrayList;
026 import java.util.Comparator;
027 import java.util.List;
028 import java.util.function.ToIntFunction;
029
030 import io.jenetics.Gene;
031 import io.jenetics.Optimize;
032 import io.jenetics.Phenotype;
033 import io.jenetics.Selector;
034 import io.jenetics.internal.math.Subset;
035 import io.jenetics.util.ISeq;
036 import io.jenetics.util.RandomRegistry;
037 import io.jenetics.util.Seq;
038
039 /**
040 * Unique fitness based tournament selection.
041 * <p>
042 * <em>The selection of unique fitnesses lifts the selection bias towards
043 * over-represented fitnesses by reducing multiple solutions sharing the same
044 * fitness to a single point in the objective space. It is therefore no longer
045 * required to assign a crowding distance of zero to individual of equal fitness
046 * as the selection operator correctly enforces diversity preservation by
047 * picking unique points in the objective space.</em>
048 * <p>
049 * <b>Reference:</b><em>
050 * Félix-Antoine Fortin and Marc Parizeau. 2013. Revisiting the NSGA-II
051 * crowding-distance computation. In Proceedings of the 15th annual
052 * conference on Genetic and evolutionary computation (GECCO '13),
053 * Christian Blum (Ed.). ACM, New York, NY, USA, 623-630.
054 * DOI=<a href="http://dx.doi.org/10.1145/2463372.2463456">
055 * 10.1145/2463372.2463456</a></em>
056 *
057 *
058 * @author <a href="mailto:franz.wilhelmstoetter@gmail.com">Franz Wilhelmstötter</a>
059 * @version 4.1
060 * @since 4.1
061 */
062 public class UFTournamentSelector<
063 G extends Gene<?, G>,
064 C extends Comparable<? super C>
065 >
066 implements Selector<G, C>
067 {
068 private final Comparator<Phenotype<G, C>> _dominance;
069 private final ElementComparator<Phenotype<G, C>> _comparator;
070 private final ElementDistance<Phenotype<G, C>> _distance;
071 private final ToIntFunction<Phenotype<G, C>> _dimension;
072
073 /**
074 * Creates a new {@code UFTournamentSelector} with the functions needed for
075 * handling the multi-objective result type {@code C}. For the {@link Vec}
076 * classes, a selector is created like in the following example:
077 * <pre>{@code
078 * new UFTournamentSelector<>(
079 * Vec<T>::dominance,
080 * Vec<T>::compare,
081 * Vec<T>::distance,
082 * Vec<T>::length
083 * );
084 * }</pre>
085 *
086 * @see #ofVec()
087 *
088 * @param dominance the pareto dominance comparator
089 * @param comparator the vector element comparator
090 * @param distance the vector element distance
091 * @param dimension the dimensionality of vector type {@code C}
092 */
093 public UFTournamentSelector(
094 final Comparator<? super C> dominance,
095 final ElementComparator<? super C> comparator,
096 final ElementDistance<? super C> distance,
097 final ToIntFunction<? super C> dimension
098 ) {
099 requireNonNull(dominance);
100 requireNonNull(comparator);
101 requireNonNull(distance);
102 requireNonNull(dimension);
103
104 _dominance = (a, b) -> dominance.compare(a.fitness(), b.fitness());
105 _comparator = comparator.map(Phenotype::fitness);
106 _distance = distance.map(Phenotype::fitness);
107 _dimension = v -> dimension.applyAsInt(v.fitness());
108 }
109
110 @Override
111 public ISeq<Phenotype<G, C>> select(
112 final Seq<Phenotype<G, C>> population,
113 final int count,
114 final Optimize opt
115 ) {
116 final var random = RandomRegistry.random();
117
118 final CrowdedComparator<Phenotype<G, C>> cc = new CrowdedComparator<>(
119 population,
120 opt,
121 _dominance,
122 _comparator,
123 _distance,
124 _dimension
125 );
126
127 final List<Phenotype<G, C>> S = new ArrayList<>();
128 while (S.size() < count) {
129 final int k = min(2*count - S.size(), population.size());
130 final int[] G = Subset.next(population.size(), k, random);
131
132 for (int j = 0; j < G.length - 1 && S.size() < count; j += 2) {
133 final int cmp = cc.compare(G[j], G[j + 1]);
134 final int p;
135 if (cmp > 0) {
136 p = G[j];
137 } else if (cmp < 0) {
138 p = G[j + 1];
139 } else {
140 p = random.nextBoolean() ? G[j] : G[j + 1];
141 }
142
143 final C fitness = population.get(p).fitness();
144 final List<Phenotype<G, C>> list = population.stream()
145 .filter(pt -> pt.fitness().equals(fitness))
146 .toList();
147
148 S.add(list.get(random.nextInt(list.size())));
149 }
150 }
151
152 return ISeq.of(S);
153 }
154
155 /**
156 * Return a new selector for the given result type {@code V}. This method is
157 * a shortcut for
158 * <pre>{@code
159 * new UFTournamentSelector<>(
160 * Vec<T>::dominance,
161 * Vec<T>::compare,
162 * Vec<T>::distance,
163 * Vec<T>::length
164 * );
165 * }</pre>
166 *
167 * @param <G> the gene type
168 * @param <T> the array type, e.g. {@code double[]}
169 * @param <V> the multi object result type vector
170 * @return a new selector for the given result type {@code V}
171 */
172 public static <G extends Gene<?, G>, T, V extends Vec<T>>
173 UFTournamentSelector<G, V> ofVec() {
174 return new UFTournamentSelector<>(
175 Vec::dominance,
176 Vec::compare,
177 Vec::distance,
178 Vec::length
179 );
180 }
181
182 }
|