[15045] | 1 | #region License Information
|
---|
| 2 | /* HeuristicLab
|
---|
| 3 | * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
| 4 | * and the BEACON Center for the Study of Evolution in Action.
|
---|
| 5 | *
|
---|
| 6 | * This file is part of HeuristicLab.
|
---|
| 7 | *
|
---|
| 8 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
| 9 | * it under the terms of the GNU General Public License as published by
|
---|
| 10 | * the Free Software Foundation, either version 3 of the License, or
|
---|
| 11 | * (at your option) any later version.
|
---|
| 12 | *
|
---|
| 13 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 16 | * GNU General Public License for more details.
|
---|
| 17 | *
|
---|
| 18 | * You should have received a copy of the GNU General Public License
|
---|
| 19 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
| 20 | */
|
---|
| 21 | #endregion
|
---|
| 22 |
|
---|
| 23 | using System;
|
---|
| 24 | using System.Collections.Generic;
|
---|
| 25 | using System.Linq;
|
---|
| 26 | using System.Threading;
|
---|
| 27 | using HeuristicLab.Analysis;
|
---|
| 28 | using HeuristicLab.Common;
|
---|
| 29 | using HeuristicLab.Core;
|
---|
| 30 | using HeuristicLab.Data;
|
---|
| 31 | using HeuristicLab.Encodings.RealVectorEncoding;
|
---|
| 32 | using HeuristicLab.Optimization;
|
---|
| 33 | using HeuristicLab.Parameters;
|
---|
| 34 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
| 35 | using HeuristicLab.Problems.TestFunctions.MultiObjective;
|
---|
| 36 | using HeuristicLab.Random;
|
---|
| 37 |
|
---|
| 38 | namespace HeuristicLab.Algorithms.MOCMAEvolutionStrategy {
|
---|
| 39 | [Item("MOCMA Evolution Strategy (MOCMAES)", "A multi objective evolution strategy based on covariance matrix adaptation. Code is based on 'Covariance Matrix Adaptation for Multi - objective Optimization' by Igel, Hansen and Roth")]
|
---|
| 40 | [Creatable(CreatableAttribute.Categories.PopulationBasedAlgorithms, Priority = 210)]
|
---|
| 41 | [StorableClass]
|
---|
| 42 | [System.Runtime.InteropServices.Guid("5AC20A69-BBBF-4153-B57D-3EAF92DC505E")]
|
---|
| 43 | public class MOCMAEvolutionStrategy : BasicAlgorithm {
|
---|
| 44 | public override Type ProblemType
|
---|
| 45 | {
|
---|
| 46 | get { return typeof(MultiObjectiveBasicProblem<RealVectorEncoding>); }
|
---|
| 47 | }
|
---|
| 48 | public new MultiObjectiveBasicProblem<RealVectorEncoding> Problem
|
---|
| 49 | {
|
---|
| 50 | get { return (MultiObjectiveBasicProblem<RealVectorEncoding>)base.Problem; }
|
---|
| 51 | set { base.Problem = value; }
|
---|
| 52 | }
|
---|
| 53 | public override bool SupportsPause
|
---|
| 54 | {
|
---|
| 55 | get { return true; }
|
---|
| 56 | }
|
---|
| 57 |
|
---|
| 58 | #region storable fields
|
---|
| 59 | [Storable]
|
---|
| 60 | private IRandom random = new MersenneTwister();
|
---|
| 61 | [Storable]
|
---|
| 62 | private NormalDistributedRandom gauss;
|
---|
| 63 | [Storable]
|
---|
| 64 | private Individual[] solutions;
|
---|
| 65 | [Storable]
|
---|
| 66 | private double stepSizeLearningRate; //=cp learning rate in [0,1]
|
---|
| 67 | [Storable]
|
---|
| 68 | private double stepSizeDampeningFactor; //d
|
---|
| 69 | [Storable]
|
---|
| 70 | private double targetSuccessProbability;// p^target_succ
|
---|
| 71 | [Storable]
|
---|
| 72 | private double evolutionPathLearningRate;//cc
|
---|
| 73 | [Storable]
|
---|
| 74 | private double covarianceMatrixLearningRate;//ccov
|
---|
| 75 | [Storable]
|
---|
| 76 | private double covarianceMatrixUnlearningRate;
|
---|
| 77 | [Storable]
|
---|
| 78 | private double successThreshold; //ptresh
|
---|
| 79 | #endregion
|
---|
| 80 |
|
---|
| 81 | #region ParameterNames
|
---|
| 82 | private const string MaximumRuntimeName = "Maximum Runtime";
|
---|
| 83 | private const string SeedName = "Seed";
|
---|
| 84 | private const string SetSeedRandomlyName = "SetSeedRandomly";
|
---|
| 85 | private const string PopulationSizeName = "PopulationSize";
|
---|
| 86 | private const string MaximumGenerationsName = "MaximumGenerations";
|
---|
| 87 | private const string MaximumEvaluatedSolutionsName = "MaximumEvaluatedSolutions";
|
---|
| 88 | private const string InitialSigmaName = "InitialSigma";
|
---|
| 89 | private const string IndicatorName = "Indicator";
|
---|
| 90 |
|
---|
| 91 | private const string EvaluationsResultName = "Evaluations";
|
---|
| 92 | private const string IterationsResultName = "Generations";
|
---|
| 93 | private const string TimetableResultName = "Timetable";
|
---|
| 94 | private const string HypervolumeResultName = "Hypervolume";
|
---|
| 95 | private const string GenerationalDistanceResultName = "Generational Distance";
|
---|
| 96 | private const string InvertedGenerationalDistanceResultName = "Inverted Generational Distance";
|
---|
| 97 | private const string CrowdingResultName = "Crowding";
|
---|
| 98 | private const string SpacingResultName = "Spacing";
|
---|
| 99 | private const string CurrentFrontResultName = "Pareto Front";
|
---|
| 100 | private const string BestHypervolumeResultName = "Best Hypervolume";
|
---|
| 101 | private const string BestKnownHypervolumeResultName = "Best known hypervolume";
|
---|
| 102 | private const string DifferenceToBestKnownHypervolumeResultName = "Absolute Distance to BestKnownHypervolume";
|
---|
| 103 | private const string ScatterPlotResultName = "ScatterPlot";
|
---|
| 104 | #endregion
|
---|
| 105 |
|
---|
| 106 | #region ParameterProperties
|
---|
| 107 | public IFixedValueParameter<IntValue> MaximumRuntimeParameter
|
---|
| 108 | {
|
---|
| 109 | get { return (IFixedValueParameter<IntValue>)Parameters[MaximumRuntimeName]; }
|
---|
| 110 | }
|
---|
| 111 | public IFixedValueParameter<IntValue> SeedParameter
|
---|
| 112 | {
|
---|
| 113 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedName]; }
|
---|
| 114 | }
|
---|
| 115 | public FixedValueParameter<BoolValue> SetSeedRandomlyParameter
|
---|
| 116 | {
|
---|
| 117 | get { return (FixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyName]; }
|
---|
| 118 | }
|
---|
| 119 | public IFixedValueParameter<IntValue> PopulationSizeParameter
|
---|
| 120 | {
|
---|
| 121 | get { return (IFixedValueParameter<IntValue>)Parameters[PopulationSizeName]; }
|
---|
| 122 | }
|
---|
| 123 | public IFixedValueParameter<IntValue> MaximumGenerationsParameter
|
---|
| 124 | {
|
---|
| 125 | get { return (IFixedValueParameter<IntValue>)Parameters[MaximumGenerationsName]; }
|
---|
| 126 | }
|
---|
| 127 | public IFixedValueParameter<IntValue> MaximumEvaluatedSolutionsParameter
|
---|
| 128 | {
|
---|
| 129 | get { return (IFixedValueParameter<IntValue>)Parameters[MaximumEvaluatedSolutionsName]; }
|
---|
| 130 | }
|
---|
| 131 | public IValueParameter<DoubleArray> InitialSigmaParameter
|
---|
| 132 | {
|
---|
| 133 | get { return (IValueParameter<DoubleArray>)Parameters[InitialSigmaName]; }
|
---|
| 134 | }
|
---|
| 135 | public IConstrainedValueParameter<IIndicator> IndicatorParameter
|
---|
| 136 | {
|
---|
| 137 | get { return (IConstrainedValueParameter<IIndicator>)Parameters[IndicatorName]; }
|
---|
| 138 | }
|
---|
| 139 | #endregion
|
---|
| 140 |
|
---|
| 141 | #region Properties
|
---|
| 142 | public int MaximumRuntime
|
---|
| 143 | {
|
---|
| 144 | get { return MaximumRuntimeParameter.Value.Value; }
|
---|
| 145 | set { MaximumRuntimeParameter.Value.Value = value; }
|
---|
| 146 | }
|
---|
| 147 | public int Seed
|
---|
| 148 | {
|
---|
| 149 | get { return SeedParameter.Value.Value; }
|
---|
| 150 | set { SeedParameter.Value.Value = value; }
|
---|
| 151 | }
|
---|
| 152 | public bool SetSeedRandomly
|
---|
| 153 | {
|
---|
| 154 | get { return SetSeedRandomlyParameter.Value.Value; }
|
---|
| 155 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
---|
| 156 | }
|
---|
| 157 | public int PopulationSize
|
---|
| 158 | {
|
---|
| 159 | get { return PopulationSizeParameter.Value.Value; }
|
---|
| 160 | set { PopulationSizeParameter.Value.Value = value; }
|
---|
| 161 | }
|
---|
| 162 | public int MaximumGenerations
|
---|
| 163 | {
|
---|
| 164 | get { return MaximumGenerationsParameter.Value.Value; }
|
---|
| 165 | set { MaximumGenerationsParameter.Value.Value = value; }
|
---|
| 166 | }
|
---|
| 167 | public int MaximumEvaluatedSolutions
|
---|
| 168 | {
|
---|
| 169 | get { return MaximumEvaluatedSolutionsParameter.Value.Value; }
|
---|
| 170 | set { MaximumEvaluatedSolutionsParameter.Value.Value = value; }
|
---|
| 171 | }
|
---|
| 172 | public DoubleArray InitialSigma
|
---|
| 173 | {
|
---|
| 174 | get { return InitialSigmaParameter.Value; }
|
---|
| 175 | set { InitialSigmaParameter.Value = value; }
|
---|
| 176 | }
|
---|
| 177 | public IIndicator Indicator
|
---|
| 178 | {
|
---|
| 179 | get { return IndicatorParameter.Value; }
|
---|
| 180 | set { IndicatorParameter.Value = value; }
|
---|
| 181 | }
|
---|
| 182 |
|
---|
| 183 | public double StepSizeLearningRate { get { return stepSizeLearningRate; } }
|
---|
| 184 | public double StepSizeDampeningFactor { get { return stepSizeDampeningFactor; } }
|
---|
| 185 | public double TargetSuccessProbability { get { return targetSuccessProbability; } }
|
---|
| 186 | public double EvolutionPathLearningRate { get { return evolutionPathLearningRate; } }
|
---|
| 187 | public double CovarianceMatrixLearningRate { get { return covarianceMatrixLearningRate; } }
|
---|
| 188 | public double CovarianceMatrixUnlearningRate { get { return covarianceMatrixUnlearningRate; } }
|
---|
| 189 | public double SuccessThreshold { get { return successThreshold; } }
|
---|
| 190 | #endregion
|
---|
| 191 |
|
---|
| 192 | #region ResultsProperties
|
---|
| 193 | private int ResultsEvaluations
|
---|
| 194 | {
|
---|
| 195 | get { return ((IntValue)Results[EvaluationsResultName].Value).Value; }
|
---|
| 196 | set { ((IntValue)Results[EvaluationsResultName].Value).Value = value; }
|
---|
| 197 | }
|
---|
| 198 | private int ResultsIterations
|
---|
| 199 | {
|
---|
| 200 | get { return ((IntValue)Results[IterationsResultName].Value).Value; }
|
---|
| 201 | set { ((IntValue)Results[IterationsResultName].Value).Value = value; }
|
---|
| 202 | }
|
---|
| 203 | #region Datatable
|
---|
| 204 | private DataTable ResultsQualities
|
---|
| 205 | {
|
---|
| 206 | get { return (DataTable)Results[TimetableResultName].Value; }
|
---|
| 207 | }
|
---|
| 208 | private DataRow ResultsBestHypervolumeDataLine
|
---|
| 209 | {
|
---|
| 210 | get { return ResultsQualities.Rows[BestHypervolumeResultName]; }
|
---|
| 211 | }
|
---|
| 212 | private DataRow ResultsHypervolumeDataLine
|
---|
| 213 | {
|
---|
| 214 | get { return ResultsQualities.Rows[HypervolumeResultName]; }
|
---|
| 215 | }
|
---|
| 216 | private DataRow ResultsGenerationalDistanceDataLine
|
---|
| 217 | {
|
---|
| 218 | get { return ResultsQualities.Rows[GenerationalDistanceResultName]; }
|
---|
| 219 | }
|
---|
| 220 | private DataRow ResultsInvertedGenerationalDistanceDataLine
|
---|
| 221 | {
|
---|
| 222 | get { return ResultsQualities.Rows[InvertedGenerationalDistanceResultName]; }
|
---|
| 223 | }
|
---|
| 224 | private DataRow ResultsCrowdingDataLine
|
---|
| 225 | {
|
---|
| 226 | get { return ResultsQualities.Rows[CrowdingResultName]; }
|
---|
| 227 | }
|
---|
| 228 | private DataRow ResultsSpacingDataLine
|
---|
| 229 | {
|
---|
| 230 | get { return ResultsQualities.Rows[SpacingResultName]; }
|
---|
| 231 | }
|
---|
| 232 | private DataRow ResultsHypervolumeDifferenceDataLine
|
---|
| 233 | {
|
---|
| 234 | get { return ResultsQualities.Rows[DifferenceToBestKnownHypervolumeResultName]; }
|
---|
| 235 | }
|
---|
| 236 | #endregion
|
---|
| 237 | //QualityIndicators
|
---|
| 238 | private double ResultsHypervolume
|
---|
| 239 | {
|
---|
| 240 | get { return ((DoubleValue)Results[HypervolumeResultName].Value).Value; }
|
---|
| 241 | set { ((DoubleValue)Results[HypervolumeResultName].Value).Value = value; }
|
---|
| 242 | }
|
---|
| 243 | private double ResultsGenerationalDistance
|
---|
| 244 | {
|
---|
| 245 | get { return ((DoubleValue)Results[GenerationalDistanceResultName].Value).Value; }
|
---|
| 246 | set { ((DoubleValue)Results[GenerationalDistanceResultName].Value).Value = value; }
|
---|
| 247 | }
|
---|
| 248 | private double ResultsInvertedGenerationalDistance
|
---|
| 249 | {
|
---|
| 250 | get { return ((DoubleValue)Results[InvertedGenerationalDistanceResultName].Value).Value; }
|
---|
| 251 | set { ((DoubleValue)Results[InvertedGenerationalDistanceResultName].Value).Value = value; }
|
---|
| 252 | }
|
---|
| 253 | private double ResultsCrowding
|
---|
| 254 | {
|
---|
| 255 | get { return ((DoubleValue)Results[CrowdingResultName].Value).Value; }
|
---|
| 256 | set { ((DoubleValue)Results[CrowdingResultName].Value).Value = value; }
|
---|
| 257 | }
|
---|
| 258 | private double ResultsSpacing
|
---|
| 259 | {
|
---|
| 260 | get { return ((DoubleValue)Results[SpacingResultName].Value).Value; }
|
---|
| 261 | set { ((DoubleValue)Results[SpacingResultName].Value).Value = value; }
|
---|
| 262 | }
|
---|
| 263 | private double ResultsBestHypervolume
|
---|
| 264 | {
|
---|
| 265 | get { return ((DoubleValue)Results[BestHypervolumeResultName].Value).Value; }
|
---|
| 266 | set { ((DoubleValue)Results[BestHypervolumeResultName].Value).Value = value; }
|
---|
| 267 | }
|
---|
| 268 | private double ResultsBestKnownHypervolume
|
---|
| 269 | {
|
---|
| 270 | get { return ((DoubleValue)Results[BestKnownHypervolumeResultName].Value).Value; }
|
---|
| 271 | set { ((DoubleValue)Results[BestKnownHypervolumeResultName].Value).Value = value; }
|
---|
| 272 | }
|
---|
| 273 | private double ResultsDifferenceBestKnownHypervolume
|
---|
| 274 | {
|
---|
| 275 | get { return ((DoubleValue)Results[DifferenceToBestKnownHypervolumeResultName].Value).Value; }
|
---|
| 276 | set { ((DoubleValue)Results[DifferenceToBestKnownHypervolumeResultName].Value).Value = value; }
|
---|
| 277 |
|
---|
| 278 | }
|
---|
| 279 | //Solutions
|
---|
| 280 | private DoubleMatrix ResultsSolutions
|
---|
| 281 | {
|
---|
| 282 | get { return (DoubleMatrix)Results[CurrentFrontResultName].Value; }
|
---|
| 283 | set { Results[CurrentFrontResultName].Value = value; }
|
---|
| 284 | }
|
---|
| 285 | private ScatterPlotContent ResultsScatterPlot
|
---|
| 286 | {
|
---|
| 287 | get { return (ScatterPlotContent)Results[ScatterPlotResultName].Value; }
|
---|
| 288 | set { Results[ScatterPlotResultName].Value = value; }
|
---|
| 289 | }
|
---|
| 290 | #endregion
|
---|
| 291 |
|
---|
| 292 | #region Constructors
|
---|
| 293 | public MOCMAEvolutionStrategy() {
|
---|
| 294 | Parameters.Add(new FixedValueParameter<IntValue>(MaximumRuntimeName, "The maximum runtime in seconds after which the algorithm stops. Use -1 to specify no limit for the runtime", new IntValue(3600)));
|
---|
| 295 | Parameters.Add(new FixedValueParameter<IntValue>(SeedName, "The random seed used to initialize the new pseudo random number generator.", new IntValue(0)));
|
---|
| 296 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyName, "True if the random seed should be set to a random value, otherwise false.", new BoolValue(true)));
|
---|
| 297 | Parameters.Add(new FixedValueParameter<IntValue>(PopulationSizeName, "λ (lambda) - the size of the offspring population.", new IntValue(20)));
|
---|
| 298 | Parameters.Add(new ValueParameter<DoubleArray>(InitialSigmaName, "The initial sigma can be a single value or a value for each dimension. All values need to be > 0.", new DoubleArray(new[] { 0.5 })));
|
---|
| 299 | Parameters.Add(new FixedValueParameter<IntValue>(MaximumGenerationsName, "The maximum number of generations which should be processed.", new IntValue(1000)));
|
---|
| 300 | Parameters.Add(new FixedValueParameter<IntValue>(MaximumEvaluatedSolutionsName, "The maximum number of evaluated solutions that should be computed.", new IntValue(int.MaxValue)));
|
---|
| 301 | var set = new ItemSet<IIndicator> { new HypervolumeIndicator(), new CrowdingIndicator(), new MinimalDistanceIndicator() };
|
---|
| 302 | Parameters.Add(new ConstrainedValueParameter<IIndicator>(IndicatorName, "The selection mechanism on non-dominated solutions", set, set.First()));
|
---|
| 303 | }
|
---|
| 304 |
|
---|
| 305 | [StorableConstructor]
|
---|
| 306 | protected MOCMAEvolutionStrategy(bool deserializing) : base(deserializing) { }
|
---|
| 307 |
|
---|
| 308 | protected MOCMAEvolutionStrategy(MOCMAEvolutionStrategy original, Cloner cloner) : base(original, cloner) {
|
---|
| 309 | random = cloner.Clone(original.random);
|
---|
| 310 | gauss = cloner.Clone(original.gauss);
|
---|
| 311 | solutions = original.solutions.Select(cloner.Clone).ToArray();
|
---|
| 312 | stepSizeLearningRate = original.stepSizeLearningRate;
|
---|
| 313 | stepSizeDampeningFactor = original.stepSizeDampeningFactor;
|
---|
| 314 | targetSuccessProbability = original.targetSuccessProbability;
|
---|
| 315 | evolutionPathLearningRate = original.evolutionPathLearningRate;
|
---|
| 316 | covarianceMatrixLearningRate = original.covarianceMatrixLearningRate;
|
---|
| 317 | covarianceMatrixUnlearningRate = original.covarianceMatrixUnlearningRate;
|
---|
| 318 | successThreshold = original.successThreshold;
|
---|
| 319 | }
|
---|
| 320 |
|
---|
| 321 | public override IDeepCloneable Clone(Cloner cloner) { return new MOCMAEvolutionStrategy(this, cloner); }
|
---|
| 322 | #endregion
|
---|
| 323 |
|
---|
| 324 | #region Initialization
|
---|
| 325 | protected override void Initialize(CancellationToken cancellationToken) {
|
---|
| 326 | if (SetSeedRandomly) Seed = new System.Random().Next();
|
---|
| 327 | random.Reset(Seed);
|
---|
| 328 | gauss = new NormalDistributedRandom(random, 0, 1);
|
---|
| 329 |
|
---|
| 330 | InitResults();
|
---|
| 331 | InitStrategy();
|
---|
| 332 | InitSolutions();
|
---|
| 333 | Analyze();
|
---|
| 334 |
|
---|
| 335 | ResultsIterations = 1;
|
---|
| 336 | cancellationToken.ThrowIfCancellationRequested();
|
---|
| 337 | }
|
---|
| 338 | private Individual InitializeIndividual(RealVector x) {
|
---|
| 339 | var zeros = new RealVector(x.Length);
|
---|
| 340 | var c = new double[x.Length, x.Length];
|
---|
| 341 | var sigma = InitialSigma.Max();
|
---|
| 342 | for (var i = 0; i < x.Length; i++) {
|
---|
| 343 | var d = InitialSigma[i % InitialSigma.Length] / sigma;
|
---|
| 344 | c[i, i] = d * d;
|
---|
| 345 | }
|
---|
| 346 | return new Individual(x, targetSuccessProbability, sigma, zeros, c, this);
|
---|
| 347 | }
|
---|
| 348 | private void InitSolutions() {
|
---|
| 349 | solutions = new Individual[PopulationSize];
|
---|
| 350 | for (var i = 0; i < PopulationSize; i++) {
|
---|
| 351 | var x = new RealVector(Problem.Encoding.Length); // Uniform distibution in all dimensions assumed.
|
---|
| 352 | var bounds = Problem.Encoding.Bounds;
|
---|
| 353 | for (var j = 0; j < Problem.Encoding.Length; j++) {
|
---|
| 354 | var dim = j % bounds.Rows;
|
---|
| 355 | x[j] = random.NextDouble() * (bounds[dim, 1] - bounds[dim, 0]) + bounds[dim, 0];
|
---|
| 356 | }
|
---|
| 357 | solutions[i] = InitializeIndividual(x);
|
---|
| 358 | PenalizeEvaluate(solutions[i]);
|
---|
| 359 | }
|
---|
| 360 | }
|
---|
| 361 | private void InitStrategy() {
|
---|
| 362 | const int lambda = 1;
|
---|
| 363 | double n = Problem.Encoding.Length;
|
---|
| 364 | targetSuccessProbability = 1.0 / (5.0 + Math.Sqrt(lambda) / 2.0);
|
---|
| 365 | stepSizeDampeningFactor = 1.0 + n / (2.0 * lambda);
|
---|
| 366 | stepSizeLearningRate = targetSuccessProbability * lambda / (2.0 + targetSuccessProbability * lambda);
|
---|
| 367 | evolutionPathLearningRate = 2.0 / (n + 2.0);
|
---|
| 368 | covarianceMatrixLearningRate = 2.0 / (n * n + 6.0);
|
---|
| 369 | covarianceMatrixUnlearningRate = 0.4 / (Math.Pow(n, 1.6) + 1);
|
---|
| 370 | successThreshold = 0.44;
|
---|
| 371 | }
|
---|
| 372 | private void InitResults() {
|
---|
| 373 | Results.Add(new Result(IterationsResultName, "The number of gererations evaluated", new IntValue(0)));
|
---|
| 374 | Results.Add(new Result(EvaluationsResultName, "The number of function evaltions performed", new IntValue(0)));
|
---|
| 375 | Results.Add(new Result(HypervolumeResultName, "The hypervolume of the current front considering the Referencepoint defined in the Problem", new DoubleValue(0.0)));
|
---|
| 376 | Results.Add(new Result(BestHypervolumeResultName, "The best hypervolume of the current run considering the Referencepoint defined in the Problem", new DoubleValue(0.0)));
|
---|
| 377 | Results.Add(new Result(BestKnownHypervolumeResultName, "The best knwon hypervolume considering the Referencepoint defined in the Problem", new DoubleValue(double.NaN)));
|
---|
| 378 | Results.Add(new Result(DifferenceToBestKnownHypervolumeResultName, "The difference between the current and the best known hypervolume", new DoubleValue(double.NaN)));
|
---|
| 379 | Results.Add(new Result(GenerationalDistanceResultName, "The generational distance to an optimal pareto front defined in the Problem", new DoubleValue(double.NaN)));
|
---|
| 380 | Results.Add(new Result(InvertedGenerationalDistanceResultName, "The inverted generational distance to an optimal pareto front defined in the Problem", new DoubleValue(double.NaN)));
|
---|
| 381 | Results.Add(new Result(CrowdingResultName, "The average crowding value for the current front (excluding infinities)", new DoubleValue(0.0)));
|
---|
| 382 | Results.Add(new Result(SpacingResultName, "The spacing for the current front (excluding infinities)", new DoubleValue(0.0)));
|
---|
| 383 |
|
---|
| 384 | var table = new DataTable("QualityIndicators");
|
---|
| 385 | table.Rows.Add(new DataRow(BestHypervolumeResultName));
|
---|
| 386 | table.Rows.Add(new DataRow(HypervolumeResultName));
|
---|
| 387 | table.Rows.Add(new DataRow(CrowdingResultName));
|
---|
| 388 | table.Rows.Add(new DataRow(GenerationalDistanceResultName));
|
---|
| 389 | table.Rows.Add(new DataRow(InvertedGenerationalDistanceResultName));
|
---|
| 390 | table.Rows.Add(new DataRow(DifferenceToBestKnownHypervolumeResultName));
|
---|
| 391 | table.Rows.Add(new DataRow(SpacingResultName));
|
---|
| 392 | Results.Add(new Result(TimetableResultName, "Different quality meassures in a timeseries", table));
|
---|
| 393 | Results.Add(new Result(CurrentFrontResultName, "The current front", new DoubleMatrix()));
|
---|
| 394 | Results.Add(new Result(ScatterPlotResultName, "A scatterplot displaying the evaluated solutions and (if available) the analytically optimal front", new ScatterPlotContent(null, null, null, 2)));
|
---|
| 395 |
|
---|
| 396 | var problem = Problem as MultiObjectiveTestFunctionProblem;
|
---|
| 397 | if (problem == null) return;
|
---|
| 398 | if (problem.BestKnownFront != null) {
|
---|
| 399 | ResultsBestKnownHypervolume = Hypervolume.Calculate(problem.BestKnownFront.ToJaggedArray(), problem.TestFunction.ReferencePoint(problem.Objectives), Problem.Maximization);
|
---|
| 400 | ResultsDifferenceBestKnownHypervolume = ResultsBestKnownHypervolume;
|
---|
| 401 | }
|
---|
| 402 | //TODO? move FrontScatterPlotContent partially? to MultiobjectiveTestProblem?
|
---|
| 403 | ResultsScatterPlot = new ScatterPlotContent(new double[0][], new double[0][], problem.BestKnownFront.ToJaggedArray(), problem.Objectives);
|
---|
| 404 | }
|
---|
| 405 | #endregion
|
---|
| 406 |
|
---|
| 407 | #region Mainloop
|
---|
| 408 | protected override void Run(CancellationToken cancellationToken) {
|
---|
| 409 | while (ResultsIterations < MaximumGenerations) {
|
---|
| 410 | try {
|
---|
| 411 | Iterate();
|
---|
| 412 | ResultsIterations++;
|
---|
| 413 | cancellationToken.ThrowIfCancellationRequested();
|
---|
| 414 | }
|
---|
| 415 | finally {
|
---|
| 416 | Analyze();
|
---|
| 417 | }
|
---|
| 418 | }
|
---|
| 419 | }
|
---|
| 420 | private void Iterate() {
|
---|
| 421 | var offspring = solutions.Select(i => {
|
---|
| 422 | var o = new Individual(i);
|
---|
| 423 | o.Mutate(gauss);
|
---|
| 424 | PenalizeEvaluate(o);
|
---|
| 425 | return o;
|
---|
| 426 | });
|
---|
| 427 | var parents = solutions.Concat(offspring).ToArray();
|
---|
| 428 | SelectParents(parents, solutions.Length);
|
---|
| 429 | UpdatePopulation(parents);
|
---|
| 430 | }
|
---|
| 431 | protected override void OnExecutionTimeChanged() {
|
---|
| 432 | base.OnExecutionTimeChanged();
|
---|
| 433 | if (CancellationTokenSource == null) return;
|
---|
| 434 | if (MaximumRuntime == -1) return;
|
---|
| 435 | if (ExecutionTime.TotalSeconds > MaximumRuntime) CancellationTokenSource.Cancel();
|
---|
| 436 | }
|
---|
| 437 | #endregion
|
---|
| 438 |
|
---|
| 439 | #region Evaluation
|
---|
| 440 | private void PenalizeEvaluate(Individual individual) {
|
---|
| 441 | if (IsFeasable(individual.Mean)) {
|
---|
| 442 | individual.Fitness = Evaluate(individual.Mean);
|
---|
| 443 | individual.PenalizedFitness = individual.Fitness;
|
---|
| 444 | } else {
|
---|
| 445 | var t = ClosestFeasible(individual.Mean);
|
---|
| 446 | individual.Fitness = Evaluate(t);
|
---|
| 447 | individual.PenalizedFitness = Penalize(individual.Mean, t, individual.Fitness);
|
---|
| 448 | }
|
---|
| 449 | }
|
---|
| 450 | private double[] Evaluate(RealVector x) {
|
---|
| 451 | var res = Problem.Evaluate(new SingleEncodingIndividual(Problem.Encoding, new Scope { Variables = { new Variable(Problem.Encoding.Name, x) } }), random);
|
---|
| 452 | ResultsEvaluations++;
|
---|
| 453 | return res;
|
---|
| 454 | }
|
---|
| 455 | private double[] Penalize(RealVector x, RealVector t, IEnumerable<double> fitness) {
|
---|
| 456 | var penalty = x.Zip(t, (a, b) => (a - b) * (a - b)).Sum() * 1E-6;
|
---|
| 457 | return fitness.Select((v, i) => Problem.Maximization[i] ? v - penalty : v + penalty).ToArray();
|
---|
| 458 | }
|
---|
| 459 | private RealVector ClosestFeasible(RealVector x) {
|
---|
| 460 | var bounds = Problem.Encoding.Bounds;
|
---|
| 461 | var r = new RealVector(x.Length);
|
---|
| 462 | for (var i = 0; i < x.Length; i++) {
|
---|
| 463 | var dim = i % bounds.Rows;
|
---|
| 464 | r[i] = Math.Min(Math.Max(bounds[dim, 0], x[i]), bounds[dim, 1]);
|
---|
| 465 | }
|
---|
| 466 | return r;
|
---|
| 467 | }
|
---|
| 468 | private bool IsFeasable(RealVector offspring) {
|
---|
| 469 | var bounds = Problem.Encoding.Bounds;
|
---|
| 470 | for (var i = 0; i < offspring.Length; i++) {
|
---|
| 471 | var dim = i % bounds.Rows;
|
---|
| 472 | if (bounds[dim, 0] > offspring[i] || offspring[i] > bounds[dim, 1]) return false;
|
---|
| 473 | }
|
---|
| 474 | return true;
|
---|
| 475 | }
|
---|
| 476 | #endregion
|
---|
| 477 |
|
---|
| 478 | private void SelectParents(IReadOnlyList<Individual> parents, int length) {
|
---|
| 479 | //perform a nondominated sort to assign the rank to every element
|
---|
| 480 | var fronts = NonDominatedSort(parents);
|
---|
| 481 |
|
---|
| 482 | //deselect the highest rank fronts until we would end up with less or equal mu elements
|
---|
| 483 | var rank = fronts.Count - 1;
|
---|
| 484 | var popSize = parents.Count;
|
---|
| 485 | while (popSize - fronts[rank].Count >= length) {
|
---|
| 486 | var front = fronts[rank];
|
---|
| 487 | foreach (var i in front) i.Selected = false;
|
---|
| 488 | popSize -= front.Count;
|
---|
| 489 | rank--;
|
---|
| 490 | }
|
---|
| 491 |
|
---|
| 492 | //now use the indicator to deselect the approximatingly worst elements of the last selected front
|
---|
| 493 | var front1 = fronts[rank].OrderBy(x => x.PenalizedFitness[0]).ToList();
|
---|
| 494 | for (; popSize > length; popSize--) {
|
---|
| 495 | var lc = Indicator.LeastContributer(front1, Problem);
|
---|
| 496 | front1[lc].Selected = false;
|
---|
| 497 | front1.Swap(lc, front1.Count - 1);
|
---|
| 498 | front1.RemoveAt(front1.Count - 1);
|
---|
| 499 | }
|
---|
| 500 | }
|
---|
| 501 |
|
---|
| 502 | private void UpdatePopulation(IReadOnlyList<Individual> parents) {
|
---|
| 503 | foreach (var p in parents.Skip(solutions.Length).Where(i => i.Selected))
|
---|
| 504 | p.UpdateAsOffspring();
|
---|
| 505 |
|
---|
| 506 | for (var i = 0; i < solutions.Length; i++)
|
---|
| 507 | if (parents[i].Selected)
|
---|
| 508 | parents[i].UpdateAsParent(parents[i + solutions.Length].Selected);
|
---|
| 509 |
|
---|
| 510 | solutions = parents.Where(p => p.Selected).ToArray();
|
---|
| 511 | }
|
---|
| 512 |
|
---|
| 513 | private void Analyze() {
|
---|
| 514 | //TODO? move FrontScatterPlotContent partially to MultiobjectiveTestProblem
|
---|
| 515 | ResultsScatterPlot = new ScatterPlotContent(solutions.Select(x => x.Fitness).ToArray(), solutions.Select(x => x.Mean.ToArray()).ToArray(), ResultsScatterPlot.ParetoFront, ResultsScatterPlot.Objectives);
|
---|
| 516 |
|
---|
| 517 | ResultsSolutions = solutions.Select(x => x.Mean.ToArray()).ToMatrix();
|
---|
| 518 |
|
---|
| 519 | var problem = Problem as MultiObjectiveTestFunctionProblem;
|
---|
| 520 | if (problem == null) return;
|
---|
| 521 |
|
---|
| 522 | var front = NonDominatedSelect.GetDominatingVectors(solutions.Select(x => x.Fitness), problem.ReferencePoint.CloneAsArray(), Problem.Maximization, true).ToArray();
|
---|
| 523 | if (front.Length == 0) return;
|
---|
| 524 | var bounds = problem.Bounds.CloneAsMatrix();
|
---|
| 525 | ResultsCrowding = Crowding.Calculate(front, bounds);
|
---|
| 526 | ResultsSpacing = Spacing.Calculate(front);
|
---|
| 527 | ResultsGenerationalDistance = problem.BestKnownFront != null ? GenerationalDistance.Calculate(front, problem.BestKnownFront.ToJaggedArray(), 1) : double.NaN;
|
---|
| 528 | ResultsInvertedGenerationalDistance = problem.BestKnownFront != null ? InvertedGenerationalDistance.Calculate(front, problem.BestKnownFront.ToJaggedArray(), 1) : double.NaN;
|
---|
| 529 | ResultsHypervolume = Hypervolume.Calculate(front, problem.ReferencePoint.CloneAsArray(), Problem.Maximization);
|
---|
| 530 | ResultsBestHypervolume = Math.Max(ResultsHypervolume, ResultsBestHypervolume);
|
---|
| 531 | ResultsDifferenceBestKnownHypervolume = ResultsBestKnownHypervolume - ResultsBestHypervolume;
|
---|
| 532 |
|
---|
| 533 | ResultsBestHypervolumeDataLine.Values.Add(ResultsBestHypervolume);
|
---|
| 534 | ResultsHypervolumeDataLine.Values.Add(ResultsHypervolume);
|
---|
| 535 | ResultsCrowdingDataLine.Values.Add(ResultsCrowding);
|
---|
| 536 | ResultsGenerationalDistanceDataLine.Values.Add(ResultsGenerationalDistance);
|
---|
| 537 | ResultsInvertedGenerationalDistanceDataLine.Values.Add(ResultsInvertedGenerationalDistance);
|
---|
| 538 | ResultsSpacingDataLine.Values.Add(ResultsSpacing);
|
---|
| 539 | ResultsHypervolumeDifferenceDataLine.Values.Add(ResultsDifferenceBestKnownHypervolume);
|
---|
| 540 |
|
---|
| 541 | Problem.Analyze(
|
---|
| 542 | solutions.Select(x => (Optimization.Individual)new SingleEncodingIndividual(Problem.Encoding, new Scope { Variables = { new Variable(Problem.Encoding.Name, x.Mean) } })).ToArray(),
|
---|
| 543 | solutions.Select(x => x.Fitness).ToArray(),
|
---|
| 544 | Results,
|
---|
| 545 | random);
|
---|
| 546 | }
|
---|
| 547 |
|
---|
| 548 | #region FastNonDominatedSort
|
---|
| 549 | //blatantly stolen form HeuristicLab.Optimization.Operators.FastNonDominatedSort
|
---|
| 550 | //however: Operators.FastNonDominatedSort does not return ranked fronts => rerank after sorting would not be sensible
|
---|
| 551 |
|
---|
| 552 | private enum DominationResult { Dominates, IsDominated, IsNonDominated };
|
---|
| 553 | private List<List<Individual>> NonDominatedSort(IReadOnlyList<Individual> individuals) {
|
---|
| 554 | const bool dominateOnEqualQualities = false;
|
---|
| 555 | var maximization = Problem.Maximization;
|
---|
| 556 | if (individuals == null) throw new InvalidOperationException(Name + ": No qualities found.");
|
---|
| 557 | var populationSize = individuals.Count;
|
---|
| 558 |
|
---|
| 559 | var fronts = new List<List<Individual>>();
|
---|
| 560 | var dominatedScopes = new Dictionary<Individual, List<int>>();
|
---|
| 561 | var dominationCounter = new int[populationSize];
|
---|
| 562 |
|
---|
| 563 | for (var pI = 0; pI < populationSize - 1; pI++) {
|
---|
| 564 | var p = individuals[pI];
|
---|
| 565 | List<int> dominatedScopesByp;
|
---|
| 566 | if (!dominatedScopes.TryGetValue(p, out dominatedScopesByp))
|
---|
| 567 | dominatedScopes[p] = dominatedScopesByp = new List<int>();
|
---|
| 568 | for (var qI = pI + 1; qI < populationSize; qI++) {
|
---|
| 569 | var test = Dominates(individuals[pI], individuals[qI], maximization, dominateOnEqualQualities);
|
---|
| 570 | if (test == DominationResult.Dominates) {
|
---|
| 571 | dominatedScopesByp.Add(qI);
|
---|
| 572 | dominationCounter[qI] += 1;
|
---|
| 573 | } else if (test == DominationResult.IsDominated) {
|
---|
| 574 | dominationCounter[pI] += 1;
|
---|
| 575 | if (!dominatedScopes.ContainsKey(individuals[qI]))
|
---|
| 576 | dominatedScopes.Add(individuals[qI], new List<int>());
|
---|
| 577 | dominatedScopes[individuals[qI]].Add(pI);
|
---|
| 578 | }
|
---|
| 579 | if (pI == populationSize - 2
|
---|
| 580 | && qI == populationSize - 1
|
---|
| 581 | && dominationCounter[qI] == 0) {
|
---|
| 582 | AddToFront(individuals[qI], fronts, 0);
|
---|
| 583 | }
|
---|
| 584 | }
|
---|
| 585 | if (dominationCounter[pI] == 0) {
|
---|
| 586 | AddToFront(p, fronts, 0);
|
---|
| 587 | }
|
---|
| 588 | }
|
---|
| 589 | var i = 0;
|
---|
| 590 | while (i < fronts.Count && fronts[i].Count > 0) {
|
---|
| 591 | var nextFront = new List<Individual>();
|
---|
| 592 | foreach (var p in fronts[i]) {
|
---|
| 593 | List<int> dominatedScopesByp;
|
---|
| 594 | if (!dominatedScopes.TryGetValue(p, out dominatedScopesByp)) continue;
|
---|
| 595 | foreach (var dominatedScope in dominatedScopesByp) {
|
---|
| 596 | dominationCounter[dominatedScope] -= 1;
|
---|
| 597 | if (dominationCounter[dominatedScope] != 0) continue;
|
---|
| 598 | nextFront.Add(individuals[dominatedScope]);
|
---|
| 599 | }
|
---|
| 600 | }
|
---|
| 601 | i += 1;
|
---|
| 602 | fronts.Add(nextFront);
|
---|
| 603 | }
|
---|
| 604 |
|
---|
| 605 | for (i = 0; i < fronts.Count; i++) {
|
---|
| 606 | foreach (var p in fronts[i]) {
|
---|
| 607 | p.Rank = i;
|
---|
| 608 | }
|
---|
| 609 | }
|
---|
| 610 | return fronts;
|
---|
| 611 | }
|
---|
| 612 | private static void AddToFront(Individual p, IList<List<Individual>> fronts, int i) {
|
---|
| 613 | if (i == fronts.Count) fronts.Add(new List<Individual>());
|
---|
| 614 | fronts[i].Add(p);
|
---|
| 615 | }
|
---|
| 616 | private static DominationResult Dominates(Individual left, Individual right, bool[] maximizations, bool dominateOnEqualQualities) {
|
---|
| 617 | return Dominates(left.PenalizedFitness, right.PenalizedFitness, maximizations, dominateOnEqualQualities);
|
---|
| 618 | }
|
---|
| 619 | private static DominationResult Dominates(IReadOnlyList<double> left, IReadOnlyList<double> right, IReadOnlyList<bool> maximizations, bool dominateOnEqualQualities) {
|
---|
| 620 | //mkommend Caution: do not use LINQ.SequenceEqual for comparing the two quality arrays (left and right) due to performance reasons
|
---|
| 621 | if (dominateOnEqualQualities) {
|
---|
| 622 | var equal = true;
|
---|
| 623 | for (var i = 0; i < left.Count; i++) {
|
---|
| 624 | if (left[i] != right[i]) {
|
---|
| 625 | equal = false;
|
---|
| 626 | break;
|
---|
| 627 | }
|
---|
| 628 | }
|
---|
| 629 | if (equal) return DominationResult.Dominates;
|
---|
| 630 | }
|
---|
| 631 |
|
---|
| 632 | bool leftIsBetter = false, rightIsBetter = false;
|
---|
| 633 | for (var i = 0; i < left.Count; i++) {
|
---|
| 634 | if (IsDominated(left[i], right[i], maximizations[i])) rightIsBetter = true;
|
---|
| 635 | else if (IsDominated(right[i], left[i], maximizations[i])) leftIsBetter = true;
|
---|
| 636 | if (leftIsBetter && rightIsBetter) break;
|
---|
| 637 | }
|
---|
| 638 |
|
---|
| 639 | if (leftIsBetter && !rightIsBetter) return DominationResult.Dominates;
|
---|
| 640 | if (!leftIsBetter && rightIsBetter) return DominationResult.IsDominated;
|
---|
| 641 | return DominationResult.IsNonDominated;
|
---|
| 642 | }
|
---|
| 643 | private static bool IsDominated(double left, double right, bool maximization) {
|
---|
| 644 | return maximization && left < right
|
---|
| 645 | || !maximization && left > right;
|
---|
| 646 | }
|
---|
| 647 | #endregion
|
---|
| 648 | }
|
---|
| 649 | }
|
---|