[17665] | 1 | using System;
|
---|
| 2 | using System.Collections.Generic;
|
---|
| 3 | using System.Linq;
|
---|
[17724] | 4 | using HEAL.Attic;
|
---|
| 5 | using HeuristicLab.Common;
|
---|
[17665] | 6 | using HeuristicLab.Core;
|
---|
| 7 | using HeuristicLab.Optimization;
|
---|
| 8 |
|
---|
| 9 | namespace HeuristicLab.Algorithms.NSGA3
|
---|
| 10 | {
|
---|
[17724] | 11 | [StorableType("53158EA5-4D1C-42DB-90F5-EC555EC0A450")]
|
---|
| 12 | public class NSGA3Selection : IDeepCloneable
|
---|
[17665] | 13 | {
|
---|
| 14 | private const double EPSILON = 10e-6; // a tiny number that is greater than 0
|
---|
| 15 |
|
---|
[17724] | 16 | [Storable]
|
---|
| 17 | private int numberOfObjectives;
|
---|
| 18 |
|
---|
| 19 | [Storable]
|
---|
| 20 | private bool[] maximization;
|
---|
| 21 |
|
---|
| 22 | [Storable]
|
---|
| 23 | private int populationSize;
|
---|
| 24 |
|
---|
| 25 | [Storable]
|
---|
| 26 | private IRandom random;
|
---|
| 27 |
|
---|
| 28 | [Storable]
|
---|
| 29 | private bool dominateOnEqualQualities;
|
---|
| 30 |
|
---|
| 31 | [StorableConstructor]
|
---|
| 32 | public NSGA3Selection(StorableConstructorFlag _) { }
|
---|
| 33 |
|
---|
| 34 | public NSGA3Selection(int numberOfObjectives, bool[] maximization, int populationSize, IRandom random, bool dominateOnEqualQualities)
|
---|
| 35 | {
|
---|
| 36 | if (maximization == null) throw new ArgumentNullException(nameof(maximization));
|
---|
| 37 | if (random == null) throw new ArgumentNullException(nameof(random));
|
---|
| 38 | if (numberOfObjectives != maximization.Length) throw new ArgumentException($"{nameof(numberOfObjectives)} and the length of {nameof(maximization)} do not match.");
|
---|
| 39 | if (populationSize < 1) throw new ArgumentOutOfRangeException($"{nameof(populationSize)} has to be >= 1.");
|
---|
| 40 |
|
---|
| 41 | this.numberOfObjectives = numberOfObjectives;
|
---|
| 42 | this.maximization = maximization;
|
---|
| 43 | this.populationSize = populationSize;
|
---|
| 44 | this.random = random;
|
---|
| 45 | this.dominateOnEqualQualities = dominateOnEqualQualities;
|
---|
| 46 | }
|
---|
| 47 |
|
---|
| 48 | public NSGA3Selection(NSGA3Selection other, Cloner cloner)
|
---|
| 49 | {
|
---|
| 50 | numberOfObjectives = other.numberOfObjectives;
|
---|
| 51 | maximization = new bool[other.maximization.Length];
|
---|
| 52 | Array.Copy(other.maximization, maximization, maximization.Length);
|
---|
| 53 | populationSize = other.populationSize;
|
---|
| 54 | random = cloner.Clone(other.random);
|
---|
| 55 | dominateOnEqualQualities = other.dominateOnEqualQualities;
|
---|
| 56 | }
|
---|
| 57 |
|
---|
| 58 | public IDeepCloneable Clone(Cloner cloner)
|
---|
| 59 | {
|
---|
| 60 | return new NSGA3Selection(this, cloner);
|
---|
| 61 | }
|
---|
| 62 |
|
---|
| 63 | public object Clone()
|
---|
| 64 | {
|
---|
| 65 | return new Cloner().Clone(this);
|
---|
| 66 | }
|
---|
| 67 |
|
---|
[17665] | 68 | /// <summary>
|
---|
[17727] | 69 | /// Select the solutions for the next generation from <see cref="rt" /> according to the
|
---|
| 70 | /// NSGA-III algorithm.
|
---|
[17665] | 71 | /// </summary>
|
---|
[17727] | 72 | /// <param name="rt">The current generation and its mutated children.</param>
|
---|
[17665] | 73 | /// <param name="referencePoints"></param>
|
---|
| 74 | /// <returns></returns>
|
---|
[17724] | 75 | public List<Solution> SelectSolutionsForNextGeneration(List<Solution> rt, List<ReferencePoint> referencePoints)
|
---|
[17665] | 76 | {
|
---|
| 77 | List<Solution> st = new List<Solution>();
|
---|
| 78 |
|
---|
| 79 | // Do non-dominated sort
|
---|
| 80 | var qualities = Utility.ToFitnessMatrix(rt);
|
---|
[17720] | 81 |
|
---|
| 82 | // compute the pareto fronts using the DominationCalculator
|
---|
[17724] | 83 | List<List<Solution>> fronts = DominationCalculator<Solution>.CalculateAllParetoFronts(rt.ToArray(), qualities, maximization, out int[] rank, dominateOnEqualQualities)
|
---|
[17665] | 84 | .Select(list => new List<Solution>(list.Select(pair => pair.Item1))).ToList();
|
---|
| 85 |
|
---|
[17727] | 86 | int last = 0; // last front to be included
|
---|
| 87 | while (st.Count < populationSize)
|
---|
[17665] | 88 | {
|
---|
[17727] | 89 | st = Utility.Concat(st, fronts[last++]);
|
---|
[17665] | 90 | }
|
---|
[17727] | 91 |
|
---|
[17668] | 92 | // remove useless fronts
|
---|
[17727] | 93 | fronts.RemoveRange(last, fronts.Count - last);
|
---|
[17665] | 94 |
|
---|
| 95 | List<Solution> nextGeneration = new List<Solution>();
|
---|
[17727] | 96 | if (st.Count == populationSize)
|
---|
| 97 | { // no special selection needs to be done
|
---|
[17665] | 98 | nextGeneration = st;
|
---|
[17727] | 99 | }
|
---|
[17665] | 100 | else
|
---|
| 101 | {
|
---|
| 102 | // Add every front to the next generation list except for the one that is added only partially
|
---|
| 103 | nextGeneration = new List<Solution>();
|
---|
[17727] | 104 | for (int f = 0; f < last - 1; f++)
|
---|
[17665] | 105 | nextGeneration = Utility.Concat(nextGeneration, fronts[f]);
|
---|
[17727] | 106 |
|
---|
[17724] | 107 | int k = populationSize - nextGeneration.Count;
|
---|
| 108 | Normalize(rt, fronts);
|
---|
[17665] | 109 | Associate(fronts, referencePoints);
|
---|
| 110 | List<Solution> solutionsToAdd = Niching(k, referencePoints, random);
|
---|
| 111 | nextGeneration = Utility.Concat(nextGeneration, solutionsToAdd);
|
---|
| 112 | }
|
---|
| 113 |
|
---|
| 114 | return nextGeneration;
|
---|
| 115 | }
|
---|
| 116 |
|
---|
[17724] | 117 | private void Normalize(List<Solution> solutions, List<List<Solution>> fronts)
|
---|
[17665] | 118 | {
|
---|
| 119 | // Find the ideal point
|
---|
[17727] | 120 | double[] idealPoint = TranslateObjectives(solutions, fronts[0]);
|
---|
[17692] | 121 |
|
---|
[17665] | 122 | // Find the extreme points
|
---|
[17727] | 123 | List<Solution> extremePoints = GetExtremePoints(fronts[0]);
|
---|
[17665] | 124 |
|
---|
| 125 | // Compute intercepts
|
---|
[17727] | 126 | double[] intercepts = GetIntercepts(extremePoints, fronts[0]);
|
---|
[17665] | 127 |
|
---|
| 128 | // Normalize objectives
|
---|
[17724] | 129 | NormalizeObjectives(solutions, intercepts, idealPoint);
|
---|
[17665] | 130 | }
|
---|
| 131 |
|
---|
[17727] | 132 | private double[] TranslateObjectives(List<Solution> solutions, List<Solution> firstFront)
|
---|
[17665] | 133 | {
|
---|
[17727] | 134 | double[] idealPoint = new double[numberOfObjectives];
|
---|
[17724] | 135 | foreach (var solution in solutions)
|
---|
[17665] | 136 | {
|
---|
[17727] | 137 | if (solution.NormalizedObjectives == null) solution.NormalizedObjectives = new double[numberOfObjectives];
|
---|
[17665] | 138 | }
|
---|
| 139 |
|
---|
[17727] | 140 | for (int j = 0; j < numberOfObjectives; j++)
|
---|
[17665] | 141 | {
|
---|
[17727] | 142 | // Compute ideal point
|
---|
| 143 | idealPoint[j] = Utility.Min(s => s.Objectives[j], firstFront);
|
---|
[17665] | 144 |
|
---|
[17727] | 145 | // Translate objectives and save the modified fitness values in NormalizedObjectives
|
---|
| 146 | foreach (var solution in solutions)
|
---|
| 147 | solution.NormalizedObjectives[j] = solution.Objectives[j] - idealPoint[j];
|
---|
[17665] | 148 | }
|
---|
| 149 |
|
---|
[17727] | 150 | return idealPoint;
|
---|
[17665] | 151 | }
|
---|
| 152 |
|
---|
[17727] | 153 | private List<Solution> GetExtremePoints(List<Solution> firstFront)
|
---|
[17665] | 154 | {
|
---|
[17727] | 155 | List<Solution> extremePoints = new List<Solution>(numberOfObjectives);
|
---|
| 156 | for (int j = 0; j < numberOfObjectives; j++)
|
---|
[17665] | 157 | {
|
---|
[17727] | 158 | // Compute extreme points
|
---|
| 159 | double[] weights = new double[numberOfObjectives];
|
---|
| 160 | for (int i = 0; i < numberOfObjectives; i++) weights[i] = EPSILON;
|
---|
| 161 | weights[j] = 1;
|
---|
[17665] | 162 |
|
---|
[17727] | 163 | var extremePoint = Utility.ArgMin(s => ASF(s.NormalizedObjectives, weights), firstFront);
|
---|
| 164 | extremePoints.Add(extremePoint);
|
---|
[17665] | 165 | }
|
---|
| 166 |
|
---|
[17727] | 167 | return extremePoints;
|
---|
[17665] | 168 | }
|
---|
| 169 |
|
---|
[17727] | 170 | private double ASF(double[] convertedFitness, double[] weight)
|
---|
[17665] | 171 | {
|
---|
[17727] | 172 | var dims = Enumerable.Range(0, numberOfObjectives);
|
---|
[17665] | 173 |
|
---|
[17727] | 174 | return Utility.Max((dim) => convertedFitness[dim] / weight[dim], dims);
|
---|
[17665] | 175 | }
|
---|
| 176 |
|
---|
[17724] | 177 | private double[] GetIntercepts(List<Solution> extremePoints, List<Solution> solutions)
|
---|
[17665] | 178 | {
|
---|
| 179 | // Check whether there are duplicate extreme points. This might happen but the original
|
---|
| 180 | // paper does not mention how to deal with it.
|
---|
| 181 | bool duplicate = false;
|
---|
| 182 | for (int i = 0; !duplicate && i < extremePoints.Count; i++)
|
---|
| 183 | {
|
---|
| 184 | for (int j = i + 1; !duplicate && j < extremePoints.Count; j++)
|
---|
| 185 | {
|
---|
[17692] | 186 | duplicate = extremePoints[i] == extremePoints[j];
|
---|
[17665] | 187 | }
|
---|
| 188 | }
|
---|
| 189 |
|
---|
[17724] | 190 | double[] intercepts = new double[numberOfObjectives];
|
---|
[17665] | 191 |
|
---|
[17724] | 192 | bool negative_intercept = false;
|
---|
| 193 | if (!duplicate)
|
---|
[17665] | 194 | {
|
---|
| 195 | // Find the equation of the hyperplane
|
---|
[17727] | 196 | List<double> b = new List<double>(Enumerable.Repeat(1.0, numberOfObjectives));
|
---|
[17665] | 197 |
|
---|
[17724] | 198 | List<List<double>> a = new List<List<double>>(extremePoints.Count);
|
---|
[17692] | 199 |
|
---|
[17724] | 200 | for (int i = 0; i < extremePoints.Count; i++)
|
---|
| 201 | {
|
---|
[17727] | 202 | List<double> convertedFitness = new List<double>(extremePoints[i].NormalizedObjectives);
|
---|
[17724] | 203 | a.Add(convertedFitness);
|
---|
| 204 | }
|
---|
[17665] | 205 | List<double> x = GaussianElimination(a, b);
|
---|
| 206 |
|
---|
| 207 | // Find intercepts
|
---|
[17724] | 208 | for (int i = 0; i < intercepts.Length; i++)
|
---|
[17665] | 209 | {
|
---|
[17724] | 210 | intercepts[i] = 1.0 / x[i];
|
---|
| 211 |
|
---|
| 212 | if (x[i] < 0)
|
---|
| 213 | {
|
---|
| 214 | negative_intercept = true;
|
---|
| 215 | break;
|
---|
| 216 | }
|
---|
[17665] | 217 | }
|
---|
| 218 | }
|
---|
| 219 |
|
---|
[17724] | 220 | // follow the method in Yuan et al. (GECCO 2015)
|
---|
| 221 | if (duplicate || negative_intercept)
|
---|
| 222 | {
|
---|
| 223 | double[] maxObjectives = FindMaxObjectives(solutions);
|
---|
| 224 | for (int i = 0; i < intercepts.Length; i++)
|
---|
| 225 | intercepts[i] = maxObjectives[i];
|
---|
| 226 | }
|
---|
| 227 |
|
---|
[17665] | 228 | return intercepts;
|
---|
| 229 | }
|
---|
| 230 |
|
---|
[17724] | 231 | private List<double> GaussianElimination(List<List<double>> a, List<double> b)
|
---|
| 232 | {
|
---|
[17665] | 233 | List<double> x = new List<double>();
|
---|
| 234 |
|
---|
| 235 | int n = a.Count;
|
---|
| 236 | for (int i = 0; i < n; i++)
|
---|
| 237 | a[i].Add(b[i]);
|
---|
| 238 |
|
---|
| 239 | for (int @base = 0; @base < n - 1; @base++)
|
---|
| 240 | for (int target = @base + 1; target < n; target++)
|
---|
| 241 | {
|
---|
| 242 | double ratio = a[target][@base] / a[@base][@base];
|
---|
| 243 | for (int term = 0; term < a[@base].Count; term++)
|
---|
| 244 | a[target][term] = a[target][term] - a[@base][term] * ratio;
|
---|
| 245 | }
|
---|
| 246 |
|
---|
| 247 | for (int i = 0; i < n; i++)
|
---|
| 248 | x.Add(0.0);
|
---|
| 249 |
|
---|
| 250 | for (int i = n - 1; i >= 0; i--)
|
---|
| 251 | {
|
---|
| 252 | for (int known = i + 1; known < n; known++)
|
---|
| 253 | a[i][n] = a[i][n] - a[i][known] * x[known];
|
---|
| 254 | x[i] = a[i][n] / a[i][i];
|
---|
| 255 | }
|
---|
| 256 |
|
---|
| 257 | return x;
|
---|
| 258 | }
|
---|
[17727] | 259 |
|
---|
| 260 | private double[] FindMaxObjectives(List<Solution> solutions)
|
---|
| 261 | {
|
---|
| 262 | double[] maxPoint = new double[numberOfObjectives];
|
---|
| 263 | for (int i = 0; i < maxPoint.Length; i++) maxPoint[i] = double.MinValue;
|
---|
| 264 |
|
---|
| 265 | for (int f = 0; f < numberOfObjectives; f++)
|
---|
| 266 | {
|
---|
| 267 | maxPoint[f] = Utility.Max(s => s.Objectives[f], solutions);
|
---|
| 268 | }
|
---|
| 269 |
|
---|
| 270 | return maxPoint;
|
---|
| 271 | }
|
---|
| 272 |
|
---|
| 273 | private void NormalizeObjectives(List<Solution> solutions, double[] intercepts, double[] idealPoint)
|
---|
| 274 | {
|
---|
| 275 | foreach (var solution in solutions)
|
---|
| 276 | {
|
---|
| 277 | for (int i = 0; i < numberOfObjectives; i++)
|
---|
| 278 | {
|
---|
| 279 | if (Math.Abs(intercepts[i] - idealPoint[i]) > 10e-10)
|
---|
| 280 | {
|
---|
| 281 | solution.NormalizedObjectives[i] = solution.NormalizedObjectives[i] / (intercepts[i] - idealPoint[i]);
|
---|
| 282 | }
|
---|
| 283 | else
|
---|
| 284 | {
|
---|
| 285 | solution.NormalizedObjectives[i] = solution.NormalizedObjectives[i] / 10e-10;
|
---|
| 286 | }
|
---|
| 287 | }
|
---|
| 288 | }
|
---|
| 289 | }
|
---|
| 290 |
|
---|
| 291 | private void Associate(List<List<Solution>> fronts, List<ReferencePoint> referencePoints)
|
---|
| 292 | {
|
---|
| 293 | for (int f = 0; f < fronts.Count; f++)
|
---|
| 294 | {
|
---|
| 295 | foreach (var solution in fronts[f])
|
---|
| 296 | {
|
---|
| 297 | // find reference point for which the perpendicular distance to the current
|
---|
| 298 | // solution is the lowest
|
---|
| 299 | var rpAndDist = Utility.MinArgMin(rp => GetPerpendicularDistance(rp.Values, solution.NormalizedObjectives), referencePoints);
|
---|
| 300 | // associated reference point
|
---|
| 301 | var arp = rpAndDist.Item1;
|
---|
| 302 | // distance to that reference point
|
---|
| 303 | var dist = rpAndDist.Item2;
|
---|
| 304 |
|
---|
| 305 | if (f + 1 != fronts.Count)
|
---|
| 306 | {
|
---|
| 307 | arp.NumberOfAssociatedSolutions++;
|
---|
| 308 | }
|
---|
| 309 | else
|
---|
| 310 | {
|
---|
| 311 | arp.AddPotentialAssociatedSolution(solution, dist);
|
---|
| 312 | }
|
---|
| 313 | }
|
---|
| 314 | }
|
---|
| 315 | }
|
---|
| 316 |
|
---|
| 317 | private double GetPerpendicularDistance(double[] direction, double[] point)
|
---|
| 318 | {
|
---|
| 319 | double numerator = 0;
|
---|
| 320 | double denominator = 0;
|
---|
| 321 | for (int i = 0; i < direction.Length; i++)
|
---|
| 322 | {
|
---|
| 323 | numerator += direction[i] * point[i];
|
---|
| 324 | denominator += Math.Pow(direction[i], 2);
|
---|
| 325 | }
|
---|
| 326 | double k = numerator / denominator;
|
---|
| 327 |
|
---|
| 328 | double d = 0;
|
---|
| 329 | for (int i = 0; i < direction.Length; i++)
|
---|
| 330 | {
|
---|
| 331 | d += Math.Pow(k * direction[i] - point[i], 2);
|
---|
| 332 | }
|
---|
| 333 | return Math.Sqrt(d);
|
---|
| 334 | }
|
---|
| 335 |
|
---|
| 336 | private List<Solution> Niching(int k, List<ReferencePoint> referencePoints, IRandom random)
|
---|
| 337 | {
|
---|
| 338 | List<Solution> solutions = new List<Solution>();
|
---|
| 339 | while (solutions.Count != k)
|
---|
| 340 | {
|
---|
| 341 | ReferencePoint min_rp = FindNicheReferencePoint(referencePoints, random);
|
---|
| 342 |
|
---|
| 343 | Solution chosen = SelectClusterMember(min_rp);
|
---|
| 344 | if (chosen == null)
|
---|
| 345 | {
|
---|
| 346 | referencePoints.Remove(min_rp);
|
---|
| 347 | }
|
---|
| 348 | else
|
---|
| 349 | {
|
---|
| 350 | min_rp.NumberOfAssociatedSolutions++;
|
---|
| 351 | min_rp.RemovePotentialAssociatedSolution(chosen);
|
---|
| 352 | solutions.Add(chosen);
|
---|
| 353 | }
|
---|
| 354 | }
|
---|
| 355 |
|
---|
| 356 | return solutions;
|
---|
| 357 | }
|
---|
| 358 |
|
---|
| 359 | private ReferencePoint FindNicheReferencePoint(List<ReferencePoint> referencePoints, IRandom random)
|
---|
| 360 | {
|
---|
| 361 | // the minimum number of associated solutions for a reference point over all reference points
|
---|
| 362 | int minNumber = Utility.Min(rp => rp.NumberOfAssociatedSolutions, referencePoints);
|
---|
| 363 |
|
---|
| 364 | var minReferencePoints = new List<ReferencePoint>(referencePoints.Where(r => r.NumberOfAssociatedSolutions == minNumber));
|
---|
| 365 | if (minReferencePoints.Count == 1) return minReferencePoints.Single();
|
---|
| 366 | else return minReferencePoints[random.Next(minReferencePoints.Count)];
|
---|
| 367 | }
|
---|
| 368 |
|
---|
| 369 | private Solution SelectClusterMember(ReferencePoint referencePoint)
|
---|
| 370 | {
|
---|
| 371 | Solution chosen = null;
|
---|
| 372 | if (referencePoint.HasPotentialMember())
|
---|
| 373 | {
|
---|
| 374 | if (referencePoint.NumberOfAssociatedSolutions == 0)
|
---|
| 375 | chosen = referencePoint.FindClosestMember();
|
---|
| 376 | else
|
---|
| 377 | chosen = referencePoint.RandomMember();
|
---|
| 378 | }
|
---|
| 379 | return chosen;
|
---|
| 380 | }
|
---|
[17665] | 381 | }
|
---|
| 382 | } |
---|