1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
4 | *
|
---|
5 | * This file is part of HeuristicLab.
|
---|
6 | *
|
---|
7 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
8 | * it under the terms of the GNU General Public License as published by
|
---|
9 | * the Free Software Foundation, either version 3 of the License, or
|
---|
10 | * (at your option) any later version.
|
---|
11 | *
|
---|
12 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
15 | * GNU General Public License for more details.
|
---|
16 | *
|
---|
17 | * You should have received a copy of the GNU General Public License
|
---|
18 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
19 | */
|
---|
20 | #endregion
|
---|
21 |
|
---|
22 | using System;
|
---|
23 | using System.Collections.Generic;
|
---|
24 | using System.Linq;
|
---|
25 | using System.Threading;
|
---|
26 | using HeuristicLab.Algorithms.MemPR.Interfaces;
|
---|
27 | using HeuristicLab.Common;
|
---|
28 | using HeuristicLab.Core;
|
---|
29 | using HeuristicLab.Encodings.BinaryVectorEncoding;
|
---|
30 | using HeuristicLab.Optimization;
|
---|
31 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
32 | using HeuristicLab.PluginInfrastructure;
|
---|
33 | using HeuristicLab.Random;
|
---|
34 |
|
---|
35 | namespace HeuristicLab.Algorithms.MemPR.Binary {
|
---|
36 | [Item("MemPR (binary)", "MemPR implementation for binary vectors.")]
|
---|
37 | [StorableClass]
|
---|
38 | [Creatable(CreatableAttribute.Categories.PopulationBasedAlgorithms, Priority = 999)]
|
---|
39 | public class BinaryMemPR : MemPRAlgorithm<SingleObjectiveBasicProblem<BinaryVectorEncoding>, BinaryVector, BinaryMemPRPopulationContext, BinaryMemPRSolutionContext> {
|
---|
40 | [StorableConstructor]
|
---|
41 | protected BinaryMemPR(bool deserializing) : base(deserializing) { }
|
---|
42 | protected BinaryMemPR(BinaryMemPR original, Cloner cloner) : base(original, cloner) { }
|
---|
43 | public BinaryMemPR() {
|
---|
44 | foreach (var trainer in ApplicationManager.Manager.GetInstances<ISolutionModelTrainer<BinaryMemPRPopulationContext>>())
|
---|
45 | SolutionModelTrainerParameter.ValidValues.Add(trainer);
|
---|
46 |
|
---|
47 | foreach (var localSearch in ApplicationManager.Manager.GetInstances<ILocalSearch<BinaryMemPRSolutionContext>>()) {
|
---|
48 | LocalSearchParameter.ValidValues.Add(localSearch);
|
---|
49 | }
|
---|
50 | }
|
---|
51 |
|
---|
52 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
53 | return new BinaryMemPR(this, cloner);
|
---|
54 | }
|
---|
55 |
|
---|
56 | protected override bool Eq(BinaryVector a, BinaryVector b) {
|
---|
57 | var len = a.Length;
|
---|
58 | for (var i = 0; i < len; i++)
|
---|
59 | if (a[i] != b[i]) return false;
|
---|
60 | return true;
|
---|
61 | }
|
---|
62 |
|
---|
63 | protected override double Dist(ISingleObjectiveSolutionScope<BinaryVector> a, ISingleObjectiveSolutionScope<BinaryVector> b) {
|
---|
64 | return 1.0 - HammingSimilarityCalculator.CalculateSimilarity(a.Solution, b.Solution);
|
---|
65 | }
|
---|
66 |
|
---|
67 | protected override ISingleObjectiveSolutionScope<BinaryVector> ToScope(BinaryVector code, double fitness = double.NaN) {
|
---|
68 | var creator = Problem.SolutionCreator as IBinaryVectorCreator;
|
---|
69 | if (creator == null) throw new InvalidOperationException("Can only solve binary encoded problems with MemPR (binary)");
|
---|
70 | return new SingleObjectiveSolutionScope<BinaryVector>(code, creator.BinaryVectorParameter.ActualName, fitness, Problem.Evaluator.QualityParameter.ActualName) {
|
---|
71 | Parent = Context.Scope
|
---|
72 | };
|
---|
73 | }
|
---|
74 |
|
---|
75 | protected override ISolutionSubspace<BinaryVector> CalculateSubspace(IEnumerable<BinaryVector> solutions, bool inverse = false) {
|
---|
76 | var pop = solutions.ToList();
|
---|
77 | var N = pop[0].Length;
|
---|
78 | var subspace = new bool[N];
|
---|
79 | for (var i = 0; i < N; i++) {
|
---|
80 | var val = pop[0][i];
|
---|
81 | if (inverse) subspace[i] = true;
|
---|
82 | for (var p = 1; p < pop.Count; p++) {
|
---|
83 | if (pop[p][i] != val) subspace[i] = !inverse;
|
---|
84 | }
|
---|
85 | }
|
---|
86 | return new BinarySolutionSubspace(subspace);
|
---|
87 | }
|
---|
88 |
|
---|
89 | protected override void AdaptiveWalk(ISingleObjectiveSolutionScope<BinaryVector> scope, int maxEvals, CancellationToken token, ISolutionSubspace<BinaryVector> subspace = null) {
|
---|
90 | var evaluations = 0;
|
---|
91 | var subset = subspace != null ? ((BinarySolutionSubspace)subspace).Subspace : null;
|
---|
92 | if (double.IsNaN(scope.Fitness)) {
|
---|
93 | Evaluate(scope, token);
|
---|
94 | evaluations++;
|
---|
95 | }
|
---|
96 | SingleObjectiveSolutionScope<BinaryVector> bestOfTheWalk = null;
|
---|
97 | var currentScope = (SingleObjectiveSolutionScope<BinaryVector>)scope.Clone();
|
---|
98 | var current = currentScope.Solution;
|
---|
99 | var N = current.Length;
|
---|
100 |
|
---|
101 | var subN = subset != null ? subset.Count(x => x) : N;
|
---|
102 | if (subN == 0) return;
|
---|
103 | var order = Enumerable.Range(0, N).Where(x => subset == null || subset[x]).Shuffle(Context.Random).ToArray();
|
---|
104 |
|
---|
105 | var bound = Problem.Maximization ? Context.Population.Max(x => x.Fitness) : Context.Population.Min(x => x.Fitness);
|
---|
106 | var range = Math.Abs(bound - Context.LocalOptimaLevel);
|
---|
107 | if (range.IsAlmost(0)) range = Math.Abs(bound * 0.05);
|
---|
108 | if (range.IsAlmost(0)) { // because bound = localoptimalevel = 0
|
---|
109 | Context.IncrementEvaluatedSolutions(evaluations);
|
---|
110 | return;
|
---|
111 | }
|
---|
112 |
|
---|
113 | var temp = -range / Math.Log(1.0 / maxEvals);
|
---|
114 | var endtemp = -range / Math.Log(0.1 / maxEvals);
|
---|
115 | var annealFactor = Math.Pow(endtemp / temp, 1.0 / maxEvals);
|
---|
116 | for (var iter = 0; iter < int.MaxValue; iter++) {
|
---|
117 | var moved = false;
|
---|
118 |
|
---|
119 | for (var i = 0; i < subN; i++) {
|
---|
120 | var idx = order[i];
|
---|
121 | var before = currentScope.Fitness;
|
---|
122 | current[idx] = !current[idx];
|
---|
123 | Evaluate(currentScope, token);
|
---|
124 | evaluations++;
|
---|
125 | var after = currentScope.Fitness;
|
---|
126 |
|
---|
127 | if (Context.IsBetter(after, before) && (bestOfTheWalk == null || Context.IsBetter(after, bestOfTheWalk.Fitness))) {
|
---|
128 | bestOfTheWalk = (SingleObjectiveSolutionScope<BinaryVector>)currentScope.Clone();
|
---|
129 | if (Context.IsBetter(bestOfTheWalk, scope)) {
|
---|
130 | moved = false;
|
---|
131 | break;
|
---|
132 | }
|
---|
133 | }
|
---|
134 | var diff = Problem.Maximization ? after - before : before - after;
|
---|
135 | if (diff > 0) moved = true;
|
---|
136 | else {
|
---|
137 | var prob = Math.Exp(diff / temp);
|
---|
138 | if (Context.Random.NextDouble() >= prob) {
|
---|
139 | // the move is not good enough -> undo the move
|
---|
140 | current[idx] = !current[idx];
|
---|
141 | currentScope.Fitness = before;
|
---|
142 | }
|
---|
143 | }
|
---|
144 | temp *= annealFactor;
|
---|
145 | if (evaluations >= maxEvals) break;
|
---|
146 | }
|
---|
147 | if (!moved) break;
|
---|
148 | if (evaluations >= maxEvals) break;
|
---|
149 | }
|
---|
150 |
|
---|
151 | Context.IncrementEvaluatedSolutions(evaluations);
|
---|
152 | scope.Adopt(bestOfTheWalk ?? currentScope);
|
---|
153 | }
|
---|
154 |
|
---|
155 | protected override ISingleObjectiveSolutionScope<BinaryVector> Breed(ISingleObjectiveSolutionScope<BinaryVector> p1, ISingleObjectiveSolutionScope<BinaryVector> p2, CancellationToken token) {
|
---|
156 | var evaluations = 0;
|
---|
157 | var N = p1.Solution.Length;
|
---|
158 | if (double.IsNaN(p1.Fitness)) {
|
---|
159 | Evaluate(p1, token);
|
---|
160 | evaluations++;
|
---|
161 | }
|
---|
162 | if (double.IsNaN(p2.Fitness)) {
|
---|
163 | Evaluate(p2, token);
|
---|
164 | evaluations++;
|
---|
165 | }
|
---|
166 | var better = Problem.Maximization ? Math.Max(p1.Fitness, p2.Fitness)
|
---|
167 | : Math.Min(p1.Fitness, p2.Fitness);
|
---|
168 |
|
---|
169 |
|
---|
170 | var cache = new HashSet<BinaryVector>(new BinaryVectorEqualityComparer());
|
---|
171 | cache.Add(p1.Solution);
|
---|
172 | cache.Add(p2.Solution);
|
---|
173 |
|
---|
174 | ISingleObjectiveSolutionScope<BinaryVector> offspring = null;
|
---|
175 | var probe = ToScope(new BinaryVector(N));
|
---|
176 | // first try all possible combinations of 1-point crossover
|
---|
177 | /*var order = Enumerable.Range(1, N - 2).Where(x => p1.Solution[x] != p2.Solution[x]).Shuffle(Context.Random).ToList();
|
---|
178 | foreach (var b in order) {
|
---|
179 | for (var i = 0; i <= b; i++)
|
---|
180 | probe.Solution[i] = p1.Solution[i];
|
---|
181 | for (var i = b + 1; i < probe.Solution.Length; i++)
|
---|
182 | probe.Solution[i] = p2.Solution[i];
|
---|
183 |
|
---|
184 | // only add to cache, because all solutions must be unique
|
---|
185 | if (cache.Contains(probe.Solution)) continue;
|
---|
186 | cache.Add(probe.Solution);
|
---|
187 | Evaluate(probe, token);
|
---|
188 | evaluations++;
|
---|
189 | if (offspring == null || Context.IsBetter(probe, offspring)) {
|
---|
190 | // immediately return upon finding a better offspring than better parent
|
---|
191 | if (Context.IsBetter(probe.Fitness, better)) {
|
---|
192 | Context.IncrementEvaluatedSolutions(evaluations);
|
---|
193 | return probe;
|
---|
194 | }
|
---|
195 | offspring = (ISingleObjectiveSolutionScope<BinaryVector>)probe.Clone();
|
---|
196 | }
|
---|
197 | }*/
|
---|
198 |
|
---|
199 | var cacheHits = 0;
|
---|
200 | // if we got some evaluations left, try uniform crossover
|
---|
201 | while (evaluations < Math.Min(Context.LocalSearchEvaluations, N)) {
|
---|
202 | probe.Solution = UniformCrossover.Apply(Context.Random, p1.Solution, p2.Solution);
|
---|
203 | if (cache.Contains(probe.Solution)) {
|
---|
204 | cacheHits++;
|
---|
205 | if (cacheHits > 10) break; // variability of uniform crossover is too low -> parents are too similar
|
---|
206 | continue;
|
---|
207 | } else cache.Add(probe.Solution);
|
---|
208 | Evaluate(probe, token);
|
---|
209 | evaluations++;
|
---|
210 | if (offspring == null || Context.IsBetter(probe, offspring)) {
|
---|
211 | // immediately return upon finding a better offspring than better parent
|
---|
212 | if (Context.IsBetter(probe.Fitness, better)) {
|
---|
213 | Context.IncrementEvaluatedSolutions(evaluations);
|
---|
214 | return probe;
|
---|
215 | }
|
---|
216 | offspring = (ISingleObjectiveSolutionScope<BinaryVector>)probe.Clone();
|
---|
217 | }
|
---|
218 | }
|
---|
219 | Context.IncrementEvaluatedSolutions(evaluations);
|
---|
220 | // return best offspring found
|
---|
221 | return offspring ?? probe;
|
---|
222 | }
|
---|
223 |
|
---|
224 | protected override ISingleObjectiveSolutionScope<BinaryVector> Link(ISingleObjectiveSolutionScope<BinaryVector> a, ISingleObjectiveSolutionScope<BinaryVector> b, CancellationToken token, bool delink = false) {
|
---|
225 | var evaluations = 0;
|
---|
226 | if (double.IsNaN(a.Fitness)) {
|
---|
227 | Evaluate(a, token);
|
---|
228 | evaluations++;
|
---|
229 | }
|
---|
230 | if (double.IsNaN(b.Fitness)) {
|
---|
231 | Evaluate(b, token);
|
---|
232 | evaluations++;
|
---|
233 | }
|
---|
234 |
|
---|
235 | var childScope = (ISingleObjectiveSolutionScope<BinaryVector>)a.Clone();
|
---|
236 | var child = childScope.Solution;
|
---|
237 | ISingleObjectiveSolutionScope<BinaryVector> best = null;
|
---|
238 | var cF = a.Fitness;
|
---|
239 | var bF = double.NaN;
|
---|
240 | var order = Enumerable.Range(0, child.Length)
|
---|
241 | .Where(x => !delink && child[x] != b.Solution[x] || delink && child[x] == b.Solution[x])
|
---|
242 | .Shuffle(Context.Random).ToList();
|
---|
243 | if (order.Count == 0) return childScope;
|
---|
244 |
|
---|
245 | while (true) {
|
---|
246 | var bestS = double.NaN;
|
---|
247 | var bestI = -1;
|
---|
248 | for (var i = 0; i < order.Count; i++) {
|
---|
249 | var idx = order[i];
|
---|
250 | child[idx] = !child[idx]; // move
|
---|
251 | Evaluate(childScope, token);
|
---|
252 | evaluations++;
|
---|
253 | var s = childScope.Fitness;
|
---|
254 | childScope.Fitness = cF;
|
---|
255 | child[idx] = !child[idx]; // undo move
|
---|
256 | if (Context.IsBetter(s, cF)) {
|
---|
257 | bestS = s;
|
---|
258 | bestI = i;
|
---|
259 | break; // first-improvement
|
---|
260 | }
|
---|
261 | if (Context.IsBetter(s, bestS)) {
|
---|
262 | // least-degrading
|
---|
263 | bestS = s;
|
---|
264 | bestI = i;
|
---|
265 | }
|
---|
266 | }
|
---|
267 | child[order[bestI]] = !child[order[bestI]];
|
---|
268 | order.RemoveAt(bestI);
|
---|
269 | cF = bestS;
|
---|
270 | childScope.Fitness = cF;
|
---|
271 | if (Context.IsBetter(cF, bF)) {
|
---|
272 | bF = cF;
|
---|
273 | best = (ISingleObjectiveSolutionScope<BinaryVector>)childScope.Clone();
|
---|
274 | }
|
---|
275 | if (order.Count == 0) break;
|
---|
276 | }
|
---|
277 | Context.IncrementEvaluatedSolutions(evaluations);
|
---|
278 | return best ?? childScope;
|
---|
279 | }
|
---|
280 | }
|
---|
281 | }
|
---|