source: branches/2936_GQAPIntegration/HeuristicLab.Problems.GeneralizedQuadraticAssignment/3.3/Operators/Crossovers/GQAPPathRelinking.cs @ 16712

Last change on this file since 16712 was 16712, checked in by gkronber, 3 months ago

#2936: adapted branch to new persistence (works with HL trunk r16711)

File size: 13.4 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2018 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
22using System;
23using System.Collections.Generic;
24using System.Linq;
25using HeuristicLab.Common;
26using HeuristicLab.Core;
27using HeuristicLab.Data;
28using HeuristicLab.Encodings.IntegerVectorEncoding;
29using HeuristicLab.Parameters;
30using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
31using HeuristicLab.Random;
32using HEAL.Attic;
33
34namespace HeuristicLab.Problems.GeneralizedQuadraticAssignment {
35  /// <summary>
36  /// This is an implementation of the algorithm described in Mateus, G.R., Resende, M.G.C. & Silva, R.M.A. J Heuristics (2011) 17: 527. https://doi.org/10.1007/s10732-010-9144-0
37  /// </summary>
38  [Item("GQAPPathRelinking", "Operator that performs path relinking between two solutions. It is described in Mateus, G., Resende, M., and Silva, R. 2011. GRASP with path-relinking for the generalized quadratic assignment problem. Journal of Heuristics 17, Springer Netherlands, pp. 527-565.")]
39  [StorableType("FAE65A24-AE6D-49DD-8A8C-6574D5304E08")]
40  public class GQAPPathRelinking : GQAPCrossover, IQualitiesAwareGQAPOperator {
41
42    public IScopeTreeLookupParameter<DoubleValue> QualityParameter {
43      get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters["Quality"]; }
44    }
45    public IScopeTreeLookupParameter<Evaluation> EvaluationParameter {
46      get { return (IScopeTreeLookupParameter<Evaluation>)Parameters["Evaluation"]; }
47    }
48    public IValueParameter<PercentValue> CandidateSizeFactorParameter {
49      get { return (IValueParameter<PercentValue>)Parameters["CandidateSizeFactor"]; }
50    }
51    public IValueLookupParameter<BoolValue> GreedyParameter {
52      get { return (IValueLookupParameter<BoolValue>)Parameters["Greedy"]; }
53    }
54
55    [StorableConstructor]
56    protected GQAPPathRelinking(StorableConstructorFlag _) : base(_) { }
57    protected GQAPPathRelinking(GQAPPathRelinking original, Cloner cloner) : base(original, cloner) { }
58    public GQAPPathRelinking()
59      : base() {
60      Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>("Quality", ""));
61      Parameters.Add(new ScopeTreeLookupParameter<Evaluation>("Evaluation", GQAP.EvaluationDescription));
62      Parameters.Add(new ValueParameter<PercentValue>("CandidateSizeFactor", "(η) Determines the size of the set of feasible moves in each path-relinking step relative to the maximum size. A value of 50% means that only half of all possible moves are considered each step.", new PercentValue(0.5)));
63      Parameters.Add(new ValueLookupParameter<BoolValue>("Greedy", "Whether to use a greedy selection strategy or a probabilistic one.", new BoolValue(true)));
64    }
65
66    public override IDeepCloneable Clone(Cloner cloner) {
67      return new GQAPPathRelinking(this, cloner);
68    }
69
70    public static GQAPSolution Apply(IRandom random,
71      IntegerVector source, Evaluation sourceEval,
72      IntegerVector target, Evaluation targetEval,
73      GQAPInstance problemInstance, double candidateSizeFactor,
74      out int evaluatedSolutions, bool greedy = true) {
75      evaluatedSolutions = 0;
76      var demands = problemInstance.Demands;
77      var capacities = problemInstance.Capacities;
78      var cmp = new IntegerVectorEqualityComparer();
79     
80      var sFit = problemInstance.ToSingleObjective(sourceEval);
81      var tFit = problemInstance.ToSingleObjective(targetEval);
82      GQAPSolution pi_star = sFit < tFit
83        ? new GQAPSolution((IntegerVector)source.Clone(), (Evaluation)sourceEval.Clone())
84        : new GQAPSolution((IntegerVector)target.Clone(), (Evaluation)targetEval.Clone()); // line 1 of Algorithm 4
85      double pi_star_Fit = problemInstance.ToSingleObjective(pi_star.Evaluation); // line 2 of Algorithm 4
86     
87      var pi_prime = (IntegerVector)source.Clone(); // line 3 of Algorithm 4
88      //var fix = new bool[demands.Length]; // line 3 of Algorithm 4, note that according to the description it is not necessary to track the fixed equipments
89      var nonFix = Enumerable.Range(0, demands.Length).ToList(); // line 3 of Algorithm 4
90      var phi = new List<int>(IntegerVectorEqualityComparer.GetDifferingIndices(pi_prime, target)); // line 4 of Algorithm 4
91
92      var B = new List<GQAPSolution>((int)Math.Ceiling(phi.Count * candidateSizeFactor));
93      var B_fit = new List<double>(B.Capacity);
94      while (phi.Count > 0) { // line 5 of Algorithm 4
95        B.Clear(); // line 6 of Algorithm 4
96        B_fit.Clear(); // line 6 of Algorithm 4 (B is split into two synchronized lists)
97        foreach (var v in phi) { // line 7 of Algorithm 4
98          int oldLocation = pi_prime[v];
99          pi_prime[v] = target[v]; // line 8 of Algorithm 4
100          var pi_dash = MakeFeasible(random, pi_prime, v, nonFix, demands, capacities); // line 9 of Algorithm 4
101          pi_prime[v] = oldLocation; // not mentioned in Algorithm 4, but seems reasonable
102
103          if (problemInstance.IsFeasible(pi_dash)) { // line 10 of Algorithm 4
104            var pi_dash_eval = problemInstance.Evaluate(pi_dash);
105            evaluatedSolutions++;
106            var pi_dash_fit = problemInstance.ToSingleObjective(pi_dash_eval);
107
108            if (B.Any(x => cmp.Equals(x.Assignment, pi_dash))) continue; // cond. 2 of line 12 and cond. 1 of line 16 in Algorithm 4
109
110            if (B.Count >= candidateSizeFactor * phi.Count) { // line 11 of Algorithm 4
111              var replacement = B_fit.Select((val, idx) => new { Index = idx, Fitness = val })
112                                            .Where(x => x.Fitness >= pi_dash_fit) // cond. 1 in line 12 of Algorithm 4
113                                            .Select(x => new { x.Index, x.Fitness, Similarity = HammingSimilarityCalculator.CalculateSimilarity(B[x.Index].Assignment, pi_dash) })
114                                            .ToArray();
115              if (replacement.Length > 0) {
116                var mostSimilar = replacement.MaxItems(x => x.Similarity).SampleRandom(random).Index;
117                B[mostSimilar].Assignment = pi_dash; // line 13 of Algorithm 4
118                B[mostSimilar].Evaluation = pi_dash_eval; // line 13 of Algorithm 4
119                B_fit[mostSimilar] = pi_dash_fit; // line 13 of Algorithm 4
120              }
121            } else { // line 16, condition has been checked above already
122              B.Add(new GQAPSolution(pi_dash, pi_dash_eval)); // line 17 of Algorithm 4
123              B_fit.Add(pi_dash_fit); // line 17 of Algorithm 4
124            }
125          }
126        }
127        if (B.Count > 0) { // line 21 of Algorithm 4
128          GQAPSolution pi;
129          // line 22 of Algorithm 4
130          if (greedy) {
131            pi = B.Select((val, idx) => new { Index = idx, Value = val }).MinItems(x => B_fit[x.Index]).SampleRandom(random).Value;
132          } else {
133            pi = B.SampleProportional(random, 1, B_fit.Select(x => 1.0 / x), false).First();
134          }
135          var diff = IntegerVectorEqualityComparer.GetDifferingIndices(pi.Assignment, target); // line 23 of Algorithm 4
136          var I = phi.Except(diff); // line 24 of Algorithm 4
137          var i = I.SampleRandom(random); // line 25 of Algorithm 4
138          //fix[i] = true; // line 26 of Algorithm 4
139          nonFix.Remove(i); // line 26 of Algorithm 4
140          pi_prime = pi.Assignment; // line 27 of Algorithm 4
141          var fit = problemInstance.ToSingleObjective(pi.Evaluation);
142          if (fit < pi_star_Fit) { // line 28 of Algorithm 4
143            pi_star_Fit = fit; // line 29 of Algorithm 4
144            pi_star = pi; // line 30 of Algorithm 4
145          }
146        } else return pi_star;
147        phi = new List<int>(IntegerVectorEqualityComparer.GetDifferingIndices(pi_prime, target));
148      }
149
150      return pi_star;
151    }
152
153    protected override IntegerVector Cross(IRandom random, ItemArray<IntegerVector> parents,
154      GQAPInstance problemInstance) {
155
156      var qualities = QualityParameter.ActualValue;
157      var evaluations = EvaluationParameter.ActualValue;
158      var betterParent = qualities[0].Value <= qualities[1].Value ? 0 : 1;
159      var worseParent = 1 - betterParent;
160      var source = parents[betterParent];
161      var target = parents[worseParent];
162
163      int evaluatedSolution;
164      return Apply(random, source, evaluations[betterParent],
165        target, evaluations[worseParent], problemInstance,
166        CandidateSizeFactorParameter.Value.Value, out evaluatedSolution,
167        GreedyParameter.ActualValue.Value).Assignment;
168    }
169
170    /// <summary>
171    /// Relocates equipments in the same location as <paramref name="equipment"/> to other locations in case the location
172    /// is overutilized.
173    /// </summary>
174    /// <remarks>
175    /// This method is performance critical, called very often and should run as fast as possible.
176    /// </remarks>
177    /// <param name="random">The random number generator.</param>
178    /// <param name="pi">The current solution.</param>
179    /// <param name="equipment">The equipment that was just assigned to a new location.</param>
180    /// <param name="nonFix">The equipments that have not yet been fixed.</param>
181    /// <param name="demands">The demands for all equipments.</param>
182    /// <param name="capacities">The capacities of all locations.</param>
183    /// <param name="maximumTries">The number of tries that should be done in relocating the equipments.</param>
184    /// <returns>A feasible or infeasible solution</returns>
185    private static IntegerVector MakeFeasible(IRandom random, IntegerVector pi, int equipment, List<int> nonFix, DoubleArray demands, DoubleArray capacities, int maximumTries = 1000) {
186      int l = pi[equipment];
187      var slack = ComputeSlack(pi, demands, capacities);
188      if (slack[l] >= 0) // line 1 of Algorithm 5
189        return (IntegerVector)pi.Clone(); // line 2 of Algorithm 5
190
191      IntegerVector pi_prime = null;
192      int k = 0; // line 4 of Algorithm 5
193      var maxSlack = slack.Max(); // line 8-9 of Algorithm 5
194      var slack_prime = (double[])slack.Clone();
195      var maxSlack_prime = maxSlack;
196      // note that FTL can be computed only once for all tries as all tries restart with the same solution
197      var FTL = nonFix.Where(x => x != equipment && pi[x] == l && demands[x] <= maxSlack).ToList(); // line 8-9 of Algorithm 5
198      var FTLweight = FTL.Select(x => demands[x]).ToList();
199      while (k < maximumTries && slack_prime[l] < 0) {  // line 5 of Algorithm 5
200        pi_prime = (IntegerVector)pi.Clone(); // line 6 of Algorithm 5
201        // set T can only shrink and not grow, thus it is created outside the loop and only updated inside
202        var T = new List<int>(FTL); // line 8-9 of Algorithm 5
203        var weightT = new List<double>(FTLweight);
204        do {  // line 7 of Algorithm 5
205          if (T.Count > 0) { // line 10 of Algorithm 5
206            var idx = Enumerable.Range(0, T.Count).SampleProportional(random, 1, weightT, false, false).First(); // line 11 of Algorithm 5
207            int i = T[idx]; // line 11 of Algorithm 5
208            var j = Enumerable.Range(0, capacities.Length)
209              .Where(x => slack_prime[x] >= demands[i]) // line 12 of Algorithm 5
210              .SampleRandom(random);  // line 13 of Algorithm 5
211            pi_prime[i] = j; // line 14 of Algorithm 5
212            T.RemoveAt(idx);
213            weightT.RemoveAt(idx);
214            var recomputeMaxSlack = slack_prime[j] == maxSlack_prime; // efficiency improvement: recompute max slack only if we assign to a location whose slack equals maxSlack
215            slack_prime[j] -= demands[i]; // line 14 of Algorithm 5
216            slack_prime[l] += demands[i]; // line 14 of Algorithm 5
217            if (recomputeMaxSlack) {
218              maxSlack_prime = slack_prime.Max();
219              // T needs to be removed of equipments whose demand is higher than maxSlack only if maxSlack changes
220              for (var h = 0; h < T.Count; h++) {
221                var f = T[h];
222                if (demands[f] > maxSlack_prime) {
223                  T.RemoveAt(h);
224                  weightT.RemoveAt(h);
225                  h--;
226                }
227              }
228            }
229          } else break; // cond. 1 in line 16 of Algorithm 5
230        } while (slack_prime[l] < 0); // cond. 2 in line 16 of Algorithm 5
231        k++; // line 17 of Algorithm 5
232        if (slack_prime[l] < 0) {
233          // reset
234          Array.Copy(slack, slack_prime, slack.Length);
235          maxSlack_prime = maxSlack;
236        }
237      }
238      return pi_prime; // line 19-23 of Algorithm 5
239    }
240
241    private static double[] ComputeSlack(IntegerVector assignment, DoubleArray demands, DoubleArray capacities) {
242      var slack = capacities.ToArray();
243      for (int i = 0; i < assignment.Length; i++) {
244        slack[assignment[i]] -= demands[i];
245      }
246      return slack;
247    }
248  }
249}
Note: See TracBrowser for help on using the repository browser.