source: branches/ichiriac/HeuristicLab.Algorithms.Shade/Shade.cs @ 15870

#region License Information
/* HeuristicLab
 * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
 *
 * This file is part of HeuristicLab.
 *
 * The implementation is inspired by the implementation in JAVA of SHADE algorithm https://sites.google.com/site/tanaberyoji/software/SHADE1.0.1_CEC2013.zip?attredirects=0&d=1
 *
 * HeuristicLab is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * HeuristicLab is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
 */

#endregion
using HeuristicLab.Analysis;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Encodings.RealVectorEncoding;
using HeuristicLab.Optimization;
using HeuristicLab.Parameters;
using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
using HeuristicLab.Problems.TestFunctions;
using HeuristicLab.Random;
using System;
using System.Collections.Generic;
using System.Threading;

namespace HeuristicLab.Algorithms.Shade {

  [Item("Success-History Based Parameter Adaptation for DE (SHADE)", "A self-adaptive version of differential evolution")]
  [StorableClass]
  [Creatable(CreatableAttribute.Categories.PopulationBasedAlgorithms, Priority = 400)]
  public class Shade : BasicAlgorithm {
    public Func<IEnumerable<double>, double> Evaluation;

    public override Type ProblemType {
      get { return typeof(SingleObjectiveTestFunctionProblem); }
    }
    public new SingleObjectiveTestFunctionProblem Problem {
      get { return (SingleObjectiveTestFunctionProblem)base.Problem; }
      set { base.Problem = value; }
    }

    private readonly IRandom _random = new MersenneTwister();
    private int evals;
    private int pop_size;
    private double arc_rate;
    private int arc_size;
    private double p_best_rate;
    private int memory_size;

    private double[][] pop;
    private double[] fitness;
    private double[][] children;
    private double[] children_fitness;

    private double[] bsf_solution;
    private double bsf_fitness = 1e+30;
    private double[,] archive;
    private int num_arc_inds = 0;

    #region ParameterNames
    private const string MaximumEvaluationsParameterName = "Maximum Evaluations";
    private const string SeedParameterName = "Seed";
    private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
    private const string CrossoverProbabilityParameterName = "CrossoverProbability";
    private const string PopulationSizeParameterName = "PopulationSize";
    private const string ScalingFactorParameterName = "ScalingFactor";
    private const string ValueToReachParameterName = "ValueToReach";
    private const string ArchiveRateParameterName = "ArchiveRate";
    private const string MemorySizeParameterName = "MemorySize";
    private const string BestRateParameterName = "BestRate";
    #endregion

    #region ParameterProperties
    public IFixedValueParameter<IntValue> MaximumEvaluationsParameter {
      get { return (IFixedValueParameter<IntValue>)Parameters[MaximumEvaluationsParameterName]; }
    }
    public IFixedValueParameter<IntValue> SeedParameter {
      get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
    }
    public FixedValueParameter<BoolValue> SetSeedRandomlyParameter {
      get { return (FixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
    }
    private ValueParameter<IntValue> PopulationSizeParameter {
      get { return (ValueParameter<IntValue>)Parameters[PopulationSizeParameterName]; }
    }
    public ValueParameter<DoubleValue> CrossoverProbabilityParameter {
      get { return (ValueParameter<DoubleValue>)Parameters[CrossoverProbabilityParameterName]; }
    }
    public ValueParameter<DoubleValue> ScalingFactorParameter {
      get { return (ValueParameter<DoubleValue>)Parameters[ScalingFactorParameterName]; }
    }
    public ValueParameter<DoubleValue> ValueToReachParameter {
      get { return (ValueParameter<DoubleValue>)Parameters[ValueToReachParameterName]; }
    }
    public ValueParameter<DoubleValue> ArchiveRateParameter {
      get { return (ValueParameter<DoubleValue>)Parameters[ArchiveRateParameterName]; }
    }
    public ValueParameter<IntValue> MemorySizeParameter {
      get { return (ValueParameter<IntValue>)Parameters[MemorySizeParameterName]; }
    }
    public ValueParameter<DoubleValue> BestRateParameter {
      get { return (ValueParameter<DoubleValue>)Parameters[BestRateParameterName]; }
    }
    #endregion

    #region Properties
    public int MaximumEvaluations {
      get { return MaximumEvaluationsParameter.Value.Value; }
      set { MaximumEvaluationsParameter.Value.Value = value; }
    }

    public Double CrossoverProbability {
      get { return CrossoverProbabilityParameter.Value.Value; }
      set { CrossoverProbabilityParameter.Value.Value = value; }
    }
    public Double ScalingFactor {
      get { return ScalingFactorParameter.Value.Value; }
      set { ScalingFactorParameter.Value.Value = value; }
    }
    public int Seed {
      get { return SeedParameter.Value.Value; }
      set { SeedParameter.Value.Value = value; }
    }
    public bool SetSeedRandomly {
      get { return SetSeedRandomlyParameter.Value.Value; }
      set { SetSeedRandomlyParameter.Value.Value = value; }
    }
    public IntValue PopulationSize {
      get { return PopulationSizeParameter.Value; }
      set { PopulationSizeParameter.Value = value; }
    }
    public Double ValueToReach {
      get { return ValueToReachParameter.Value.Value; }
      set { ValueToReachParameter.Value.Value = value; }
    }
    public Double ArchiveRate {
      get { return ArchiveRateParameter.Value.Value; }
      set { ArchiveRateParameter.Value.Value = value; }
    }
    public IntValue MemorySize {
      get { return MemorySizeParameter.Value; }
      set { MemorySizeParameter.Value = value; }
    }
    public Double BestRate {
      get { return BestRateParameter.Value.Value; }
      set { BestRateParameter.Value.Value = value; }
    }
    #endregion

    #region ResultsProperties
    private double ResultsBestQuality {
      get { return ((DoubleValue)Results["Best Quality"].Value).Value; }
      set { ((DoubleValue)Results["Best Quality"].Value).Value = value; }
    }

    private double VTRBestQuality {
      get { return ((DoubleValue)Results["VTR"].Value).Value; }
      set { ((DoubleValue)Results["VTR"].Value).Value = value; }
    }

    private RealVector ResultsBestSolution {
      get { return (RealVector)Results["Best Solution"].Value; }
      set { Results["Best Solution"].Value = value; }
    }

    private int ResultsEvaluations {
      get { return ((IntValue)Results["Evaluations"].Value).Value; }
      set { ((IntValue)Results["Evaluations"].Value).Value = value; }
    }
    private int ResultsIterations {
      get { return ((IntValue)Results["Iterations"].Value).Value; }
      set { ((IntValue)Results["Iterations"].Value).Value = value; }
    }

    private DataTable ResultsQualities {
      get { return ((DataTable)Results["Qualities"].Value); }
    }
    private DataRow ResultsQualitiesBest {
      get { return ResultsQualities.Rows["Best Quality"]; }
    }

    #endregion

    [StorableConstructor]
    protected Shade(bool deserializing) : base(deserializing) { }

    protected Shade(Shade original, Cloner cloner)
      : base(original, cloner) {
    }

    public override IDeepCloneable Clone(Cloner cloner) {
      return new Shade(this, cloner);
    }

    public Shade() {
      Parameters.Add(new FixedValueParameter<IntValue>(MaximumEvaluationsParameterName, "", new IntValue(Int32.MaxValue)));
      Parameters.Add(new ValueParameter<IntValue>(PopulationSizeParameterName, "The size of the population of solutions.", new IntValue(75)));
      Parameters.Add(new ValueParameter<DoubleValue>(ValueToReachParameterName, "Value to reach (VTR) parameter", new DoubleValue(0.00000001)));
      Parameters.Add(new ValueParameter<DoubleValue>(ArchiveRateParameterName, "Archive rate parameter", new DoubleValue(2.0)));
      Parameters.Add(new ValueParameter<IntValue>(MemorySizeParameterName, "Memory size parameter", new IntValue(0)));
      Parameters.Add(new ValueParameter<DoubleValue>(BestRateParameterName, "Best rate parameter", new DoubleValue(0.1)));
    }

    protected override void Run(CancellationToken cancellationToken) {

      // Set up the results display
      Results.Add(new Result("Iterations", new IntValue(0)));
      Results.Add(new Result("Evaluations", new IntValue(0)));
      Results.Add(new Result("Best Solution", new RealVector()));
      Results.Add(new Result("Best Quality", new DoubleValue(double.NaN)));
      Results.Add(new Result("VTR", new DoubleValue(double.NaN)));
      var table = new DataTable("Qualities");
      table.Rows.Add(new DataRow("Best Quality"));
      Results.Add(new Result("Qualities", table));


      this.evals = 0;
      int archive_size = (int)Math.Round(ArchiveRateParameter.Value.Value * PopulationSize.Value);
      int problem_size = Problem.ProblemSize.Value;

      int pop_size = PopulationSizeParameter.Value.Value;
      this.arc_rate = ArchiveRateParameter.Value.Value;
      this.arc_size = (int)Math.Round(this.arc_rate * pop_size);
      this.p_best_rate = BestRateParameter.Value.Value;
      this.memory_size = MemorySizeParameter.Value.Value;

      this.pop = new double[pop_size][];
      this.fitness = new double[pop_size];
      this.children = new double[pop_size][];
      this.children_fitness = new double[pop_size];

      this.bsf_solution = new double[problem_size];
      this.bsf_fitness = 1e+30;
      this.archive = new double[arc_size, Problem.ProblemSize.Value];
      this.num_arc_inds = 0;

      double[,] populationOld = new double[PopulationSizeParameter.Value.Value, Problem.ProblemSize.Value];
      double[,] mutationPopulation = new double[PopulationSizeParameter.Value.Value, Problem.ProblemSize.Value];
      double[,] trialPopulation = new double[PopulationSizeParameter.Value.Value, Problem.ProblemSize.Value];
      double[] bestPopulation = new double[Problem.ProblemSize.Value];
      double[] bestPopulationIteration = new double[Problem.ProblemSize.Value];
      double[,] archive = new double[archive_size, Problem.ProblemSize.Value];


      // //for external archive
      int rand_arc_ind;

      int num_success_params;

      double[] success_sf = new double[PopulationSizeParameter.Value.Value];
      double[] success_cr = new double[PopulationSizeParameter.Value.Value];
      double[] dif_fitness = new double[PopulationSizeParameter.Value.Value];
      double[] fitness = new double[PopulationSizeParameter.Value.Value];

      // the contents of M_f and M_cr are all initialiezed 0.5
      double[] memory_sf = new double[MemorySizeParameter.Value.Value];
      double[] memory_cr = new double[MemorySizeParameter.Value.Value];

      for(int i = 0; i < MemorySizeParameter.Value.Value; i++) {
        memory_sf[i] = 0.5;
        memory_cr[i] = 0.5;
      }

      //memory index counter
      int memory_pos = 0;
      double temp_sum_sf1, temp_sum_sf2, temp_sum_cr1, temp_sum_cr2, temp_sum, temp_weight;

      //for new parameters sampling
      double mu_sf, mu_cr;
      int rand_mem_index;

      double[] pop_sf = new double[PopulationSizeParameter.Value.Value];
      double[] pop_cr = new double[PopulationSizeParameter.Value.Value];

      //for current-to-pbest/1
      int p_best_ind;
      double m = PopulationSizeParameter.Value.Value * BestRateParameter.Value.Value;
      int p_num = (int)Math.Round(m);
      int[] sorted_array = new int[PopulationSizeParameter.Value.Value];
      double[] sorted_fitness = new double[PopulationSizeParameter.Value.Value];

      //initialize the population
      populationOld = makeNewIndividuals();

      //evaluate the best member after the intialiazation
      //the idea is to select first member and after that to check the others members from the population

      int best_index = 0;
      double[] populationRow = new double[Problem.ProblemSize.Value];
      bestPopulation = getMatrixRow(populationOld, best_index);
      RealVector bestPopulationVector = new RealVector(bestPopulation);
      double bestPopulationValue = Obj(bestPopulationVector);
      fitness[best_index] = bestPopulationValue;
      RealVector selectionVector;
      RealVector trialVector;
      double qtrial;


      for(var i = 0; i < PopulationSizeParameter.Value.Value; i++) {
        populationRow = getMatrixRow(populationOld, i);
        trialVector = new RealVector(populationRow);

        qtrial = Obj(trialVector);
        fitness[i] = qtrial;

        if(qtrial > bestPopulationValue) {
          bestPopulationVector = new RealVector(populationRow);
          bestPopulationValue = qtrial;
          best_index = i;
        }
      }

      int iterations = 1;

      // Loop until iteration limit reached or canceled.
      // todo replace with a function
      // && bestPopulationValue > Problem.BestKnownQuality.Value + ValueToReachParameter.Value.Value
      while(ResultsEvaluations < MaximumEvaluations
          && !cancellationToken.IsCancellationRequested &&
          bestPopulationValue > Problem.BestKnownQuality.Value + ValueToReachParameter.Value.Value) {
        for(int i = 0; i < PopulationSizeParameter.Value.Value; i++) sorted_array[i] = i;
        for(int i = 0; i < PopulationSizeParameter.Value.Value; i++) sorted_fitness[i] = fitness[i];

        Quicksort(sorted_fitness, 0, PopulationSizeParameter.Value.Value - 1, sorted_array);

        for(int target = 0; target < PopulationSizeParameter.Value.Value; target++) {
          rand_mem_index = (int)(_random.NextDouble() * MemorySizeParameter.Value.Value);
          mu_sf = memory_sf[rand_mem_index];
          mu_cr = memory_cr[rand_mem_index];

          //generate CR_i and repair its value
          if(mu_cr == -1) {
            pop_cr[target] = 0;
          } else {
            pop_cr[target] = gauss(mu_cr, 0.1);
            if(pop_cr[target] > 1) pop_cr[target] = 1;
            else if(pop_cr[target] < 0) pop_cr[target] = 0;
          }

          //generate F_i and repair its value
          do {
            pop_sf[target] = cauchy_g(mu_sf, 0.1);
          } while(pop_sf[target] <= 0);

          if(pop_sf[target] > 1) pop_sf[target] = 1;

          //p-best individual is randomly selected from the top pop_size *  p_i members
          p_best_ind = sorted_array[(int)(_random.NextDouble() * p_num)];

          trialPopulation = operateCurrentToPBest1BinWithArchive(populationOld, trialPopulation, target, p_best_ind, pop_sf[target], pop_cr[target]);
        }

        for(int i = 0; i < pop_size; i++) {
          trialVector = new RealVector(getMatrixRow(trialPopulation, i));
          children_fitness[i] = Obj(trialVector);
        }

        //update bfs solution 
        for(var i = 0; i < PopulationSizeParameter.Value.Value; i++) {
          populationRow = getMatrixRow(populationOld, i);
          qtrial = fitness[i];

          if(qtrial > bestPopulationValue) {
            bestPopulationVector = new RealVector(populationRow);
            bestPopulationValue = qtrial;
            best_index = i;
          }
        }

        num_success_params = 0;

        //generation alternation
        for(int i = 0; i < pop_size; i++) {
          if(children_fitness[i] == fitness[i]) {
            fitness[i] = children_fitness[i];
            for(int j = 0; j < problem_size; j++) populationOld[i, j] = trialPopulation[i, j];
          } else if(children_fitness[i] < fitness[i]) {
            //parent vectors x_i which were worse than the trial vectors u_i are preserved
            if(arc_size > 1) {
              if(num_arc_inds < arc_size) {
                for(int j = 0; j < problem_size; j++) this.archive[num_arc_inds, j] = populationOld[i, j];
                num_arc_inds++;

              }
              //Whenever the size of the archive exceeds, randomly selected elements are deleted to make space for the newly inserted elements
              else {
                rand_arc_ind = (int)(_random.NextDouble() * arc_size);
                for(int j = 0; j < problem_size; j++) this.archive[rand_arc_ind, j] = populationOld[i, j];
              }
            }

            dif_fitness[num_success_params] = Math.Abs(fitness[i] - children_fitness[i]);

            fitness[i] = children_fitness[i];
            for(int j = 0; j < problem_size; j++) populationOld[i, j] = trialPopulation[i, j];

            //successful parameters are preserved in S_F and S_CR
            success_sf[num_success_params] = pop_sf[i];
            success_cr[num_success_params] = pop_cr[i];
            num_success_params++;
          }
        }

        if(num_success_params > 0) {
          temp_sum_sf1 = 0;
          temp_sum_sf2 = 0;
          temp_sum_cr1 = 0;
          temp_sum_cr2 = 0;
          temp_sum = 0;
          temp_weight = 0;

          for(int i = 0; i < num_success_params; i++) temp_sum += dif_fitness[i];

          //weighted lehmer mean
          for(int i = 0; i < num_success_params; i++) {
            temp_weight = dif_fitness[i] / temp_sum;

            temp_sum_sf1 += temp_weight * success_sf[i] * success_sf[i];
            temp_sum_sf2 += temp_weight * success_sf[i];

            temp_sum_cr1 += temp_weight * success_cr[i] * success_cr[i];
            temp_sum_cr2 += temp_weight * success_cr[i];
          }

          memory_sf[memory_pos] = temp_sum_sf1 / temp_sum_sf2;

          if(temp_sum_cr2 == 0 || memory_cr[memory_pos] == -1) {
            memory_cr[memory_pos] = -1;
          } else {
            memory_cr[memory_pos] = temp_sum_cr1 / temp_sum_cr2;
          }

          //increment the counter
          memory_pos++;
          if(memory_pos >= memory_size) memory_pos = 0;
        }

        //update the best candidate
        for(int i = 0; i < PopulationSizeParameter.Value.Value; i++) {
          selectionVector = new RealVector(getMatrixRow(populationOld, i));
          var quality = fitness[i];
          if(quality < bestPopulationValue) {
            bestPopulationVector = (RealVector)selectionVector.Clone();
            bestPopulationValue = quality;
          }
        }

        iterations = iterations + 1;

        //update the results
        ResultsEvaluations = evals;
        ResultsIterations = iterations;
        ResultsBestSolution = bestPopulationVector;
        ResultsBestQuality = bestPopulationValue;

        //update the results in view
        if(iterations % 10 == 0) ResultsQualitiesBest.Values.Add(bestPopulationValue);
        if(bestPopulationValue < Problem.BestKnownQuality.Value + ValueToReachParameter.Value.Value) {
          VTRBestQuality = bestPopulationValue;
        }
      }
    }

    public override bool SupportsPause { get { return false; } } // TODO (can we pause?)

    //evaluate the vector
    public double Obj(RealVector x) {
      evals = evals + 1;
      if(Problem.Maximization.Value)
        return -Problem.Evaluator.Evaluate(x);

      return Problem.Evaluator.Evaluate(x);
    }

    // Get ith row from the matrix
    public double[] getMatrixRow(double[,] Mat, int i) {
      double[] tmp = new double[Mat.GetUpperBound(1) + 1];

      for(int j = 0; j <= Mat.GetUpperBound(1); j++) {
        tmp[j] = Mat[i, j];
      }

      return tmp;
    }

    /*
        Return random value from Cauchy distribution with mean "mu" and variance "gamma"
        http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html#Cauchy
    */
    private double cauchy_g(double mu, double gamma) {
      return mu + gamma * Math.Tan(Math.PI * (_random.NextDouble() - 0.5));
    }

    /*
         Return random value from normal distribution with mean "mu" and variance "gamma"
         http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html#Gauss
    */
    private double gauss(double mu, double sigma) {
      return mu + sigma * Math.Sqrt(-2.0 * Math.Log(_random.NextDouble())) * Math.Sin(2.0 * Math.PI * _random.NextDouble());
    }

    private double[,] makeNewIndividuals() {
      //problem variables
      var dim = Problem.ProblemSize.Value;
      var lb = Problem.Bounds[0, 0];
      var ub = Problem.Bounds[0, 1];
      var range = ub - lb;
      double[,] population = new double[PopulationSizeParameter.Value.Value, Problem.ProblemSize.Value];

      //create initial population
      //population is a matrix of size PopulationSize*ProblemSize
      for(int i = 0; i < PopulationSizeParameter.Value.Value; i++) {
        for(int j = 0; j < Problem.ProblemSize.Value; j++) {
          population[i, j] = _random.NextDouble() * range + lb;
        }
      }
      return population;
    }

    private static void Quicksort(double[] elements, int left, int right, int[] index) {
      int i = left, j = right;
      double pivot = elements[(left + right) / 2];
      double tmp_var = 0;
      int tmp_index = 0;

      while(i <= j) {
        while(elements[i].CompareTo(pivot) < 0) {
          i++;
        }

        while(elements[j].CompareTo(pivot) > 0) {
          j--;
        }

        if(i <= j) {
          // Swap
          tmp_var = elements[i];
          elements[i] = elements[j];
          elements[j] = tmp_var;

          tmp_index = index[i];
          index[i] = index[j];
          index[j] = tmp_index;

          i++;
          j--;
        }
      }

      // Recursive calls
      if(left < j) {
        Quicksort(elements, left, j, index);
      }

      if(i < right) {
        Quicksort(elements, i, right, index);
      }
    }

    // current to best selection scheme with archive
    // analyze how the archive is implemented
    private double[,] operateCurrentToPBest1BinWithArchive(double[,] pop, double[,] children, int target, int p_best_individual, double scaling_factor, double cross_rate) {
      int r1, r2;
      int num_arc_inds = 0;
      var lb = Problem.Bounds[0, 0];
      var ub = Problem.Bounds[0, 1];

      do {
        r1 = (int)(_random.NextDouble() * PopulationSizeParameter.Value.Value);
      } while(r1 == target);
      do {
        r2 = (int)(_random.NextDouble() * (PopulationSizeParameter.Value.Value + num_arc_inds));
      } while((r2 == target) || (r2 == r1));

      int random_variable = (int)(_random.NextDouble() * Problem.ProblemSize.Value);

      if(r2 >= PopulationSizeParameter.Value.Value) {
        r2 -= PopulationSizeParameter.Value.Value;
        for(int i = 0; i < Problem.ProblemSize.Value; i++) {
          if((_random.NextDouble() < cross_rate) || (i == random_variable)) children[target, i] = pop[target, i] + scaling_factor * (pop[p_best_individual, i] - pop[target, i]) + scaling_factor * (pop[r1, i] - archive[r2, i]);
          else children[target, i] = pop[target, i];
        }
      } else {
        for(int i = 0; i < Problem.ProblemSize.Value; i++) {
          if((_random.NextDouble() < cross_rate) || (i == random_variable)) children[target, i] = pop[target, i] + scaling_factor * (pop[p_best_individual, i] - pop[target, i]) + scaling_factor * (pop[r1, i] - pop[r2, i]);
          else children[target, i] = pop[target, i];
        }
      }

      for(int i = 0; i < Problem.ProblemSize.Value; i++) {
        if(children[target, i] < lb) children[target, i] = (lb + pop[target, i]) / 2.0;
        else if(children[target, i] > ub) children[target, i] = (ub + pop[target, i]) / 2.0;
      }

      return children;
    }
  }
}