public override void Evaluate(Solution solution)
{
_counter++;
IntergenSolution s = (IntergenSolution)solution;
s.FoundAtEval = _counter;
var doubleVal = new double[Model.Setting.NumberOfFeatures];
var values = new XReal(solution);
for (var i = 0; i < Model.Setting.NumberOfFeatures; i++)
{
doubleVal[i] = values.GetValue(i);
}
//the distribution from the NSGA2
var nsgaFv = new Distribution(doubleVal);
var fc = new FitnessCalculator(Model, _counter);
var fitnessValues = fc.Calculate(null, null, nsgaFv, FeatureTarget, null, null);
solution.Objective[0] = fitnessValues.FeatureVal;
FitnessTracker.AddFeat(s.FoundAtEval, fitnessValues.FeatureVal);
}
/// <summary>
/// Runs the NSGA-II algorithm.
/// </summary>
/// <returns>a <code>SolutionSet</code> that is a set of non dominated solutions as a result of the algorithm execution</returns>
public override SolutionSet Execute()
{
int populationSize = -1;
int maxEvaluations = -1;
int evaluations;
JMetalCSharp.QualityIndicator.QualityIndicator indicators = null; // QualityIndicator object
int requiredEvaluations; // Use in the example of use of the
// indicators object (see below)
SolutionSet population;
SolutionSet offspringPopulation;
SolutionSet union;
Operator mutationOperator;
Operator crossoverOperator;
Operator selectionOperator;
Distance distance = new Distance();
//Read the parameters
JMetalCSharp.Utils.Utils.GetIntValueFromParameter(this.InputParameters, "maxEvaluations", ref maxEvaluations);
JMetalCSharp.Utils.Utils.GetIntValueFromParameter(this.InputParameters, "populationSize", ref populationSize);
JMetalCSharp.Utils.Utils.GetIndicatorsFromParameters(this.InputParameters, "indicators", ref indicators);
parallelEvaluator.StartParallelRunner(Problem);;
//Initialize the variables
population = new SolutionSet(populationSize);
evaluations = 0;
requiredEvaluations = 0;
//Read the operators
mutationOperator = Operators["mutation"];
crossoverOperator = Operators["crossover"];
selectionOperator = Operators["selection"];
// Create the initial solutionSet
IntergenSolution newSolution;
for (int i = 0; i < populationSize; i++)
{
newSolution = new IntergenSolution((IntergenProblem)Problem);
parallelEvaluator.AddTaskForExecution(new object[] { newSolution, i });;
}
List <IntergenSolution> solutionList = (List <IntergenSolution>)parallelEvaluator.ParallelExecution();
foreach (IntergenSolution solution in solutionList)
{
population.Add(solution);
evaluations++;
}
// Generations
while (evaluations < maxEvaluations)
{
// Create the offSpring solutionSet
offspringPopulation = new SolutionSet(populationSize);
IntergenSolution[] parents = new IntergenSolution[2];
for (int i = 0; i < (populationSize / 2); i++)
{
if (evaluations < maxEvaluations)
{
//obtain parents
parents[0] = (IntergenSolution)selectionOperator.Execute(population);
parents[1] = (IntergenSolution)selectionOperator.Execute(population);
IntergenSolution[] offSpring = (IntergenSolution[])crossoverOperator.Execute(parents);
mutationOperator.Execute(offSpring[0]);
mutationOperator.Execute(offSpring[1]);
parallelEvaluator.AddTaskForExecution(new object[] { offSpring[0], evaluations + i });
parallelEvaluator.AddTaskForExecution(new object[] { offSpring[1], evaluations + i });
}
}
List <IntergenSolution> solutions = (List <IntergenSolution>)parallelEvaluator.ParallelExecution();
foreach (IntergenSolution solution in solutions)
{
offspringPopulation.Add(solution);
evaluations++;
//solution.FoundAtEval = evaluations;
}
// Create the solutionSet union of solutionSet and offSpring
union = ((SolutionSet)population).Union(offspringPopulation);
// Ranking the union
Ranking ranking = new Ranking(union);
int remain = populationSize;
int index = 0;
SolutionSet front = null;
population.Clear();
// Obtain the next front
front = ranking.GetSubfront(index);
while ((remain > 0) && (remain >= front.Size()))
{
//Assign crowding distance to individuals
distance.CrowdingDistanceAssignment(front, Problem.NumberOfObjectives);
//Add the individuals of this front
for (int k = 0; k < front.Size(); k++)
{
population.Add(front.Get(k));
}
//Decrement remain
remain = remain - front.Size();
//Obtain the next front
index++;
if (remain > 0)
{
front = ranking.GetSubfront(index);
}
}
// Remain is less than front(index).size, insert only the best one
if (remain > 0)
{ // front contains individuals to insert
distance.CrowdingDistanceAssignment(front, Problem.NumberOfObjectives);
front.Sort(new CrowdingComparator());
for (int k = 0; k < remain; k++)
{
population.Add(front.Get(k));
}
remain = 0;
}
// This piece of code shows how to use the indicator object into the code
// of NSGA-II. In particular, it finds the number of evaluations required
// by the algorithm to obtain a Pareto front with a hypervolume higher
// than the hypervolume of the true Pareto front.
if ((indicators != null) &&
(requiredEvaluations == 0))
{
double HV = indicators.GetHypervolume(population);
if (HV >= (0.98 * indicators.TrueParetoFrontHypervolume))
{
requiredEvaluations = evaluations;
}
}
//TODO
/*
* Ranking rank2 = new Ranking(population);
*
* Result = rank2.GetSubfront(0);
*/
/*Ranking forGraphicOutput = new Ranking(population);
* var currentBestResultSet = forGraphicOutput.GetSubfront(0);
*
* var firstBestResult = currentBestResultSet.Get(0);
*
* var myProblem = (IntergenProblem)Problem;
* //myProblem.calculated;
*
* //var variantValuesWithoutInteraction = Matrix.Multiply(myProblem.calculated, firstBestResult.);
* //var Model = myProblem.GetModel();
* int mycounter = 0;
* if (mycounter % 500 == 0)
* {
* //RIntegrator.PlotValues(currentBestResultSet, myProblem);
* }
* mycounter++;
* var progress = new UserProgress();
* progress.FeatureP = firstBestResult.Objective[0];
* if (!myProblem.Model.Setting.NoVariantCalculation) progress.VariantP = firstBestResult.Objective[1];
* myProblem.Worker.ReportProgress(evaluations * 100 / maxEvaluations, progress); ;
*
* //Model.CurrentBestImage = "CurrentBest.png"; */
front = ranking.GetSubfront(0);
var minmax = new MinMaxFitness
{
FeatMax = double.MinValue,
FeatMin = double.MaxValue,
VarMax = double.MinValue,
VarMin = double.MaxValue,
InterMax = double.MinValue,
InterMin = double.MaxValue
};
var prob = (IntergenProblem)Problem;
var list = ObjectiveMapping.GetList(prob.ProblemType);
for (var i = 0; i < populationSize; i++)
{
var sol = population.Get(i);
var objindex = 0;
if (list[0])
{
if (sol.Objective[objindex] < minmax.FeatMin)
{
minmax.FeatMin = sol.Objective[objindex];
}
if (sol.Objective[objindex] > minmax.FeatMax)
{
minmax.FeatMax = sol.Objective[objindex];
}
objindex++;
}
if (list[1])
{
if (sol.Objective[objindex] < minmax.InterMin)
{
minmax.InterMin = sol.Objective[objindex];
}
if (sol.Objective[objindex] > minmax.InterMax)
{
minmax.InterMax = sol.Objective[objindex];
}
objindex++;
}
if (list[2])
{
if (sol.Objective[objindex] < minmax.VarMin)
{
minmax.VarMin = sol.Objective[objindex];
}
if (sol.Objective[objindex] > minmax.VarMax)
{
minmax.VarMax = sol.Objective[objindex];
}
}
}
var sol0 = front.Best(new CrowdingDistanceComparator());
var done = FitnessTracker.AddFitn(minmax);
SolutionPlotter.Plot(sol0);
ProgressReporter.ReportSolution(evaluations, sol0, _worker);
if (done)
{
Ranking rank3 = new Ranking(population);
Result = rank3.GetSubfront(0);
SetOutputParameter("evaluations", evaluations);
return(this.Result);
}
}
// Return as output parameter the required evaluations
SetOutputParameter("evaluations", evaluations);
// Return the first non-dominated front
Ranking rank = new Ranking(population);
Result = rank.GetSubfront(0);
return(this.Result);
}
public override void Evaluate(Solution solution)
{
#if DEBUG
if (_counter == 0)
{
if (Model.Setting.Logging)
{
LogArrayValues(FeatureTarget.Values, "targetFeatures");
if (Model.Setting.NumberOfInteractions > 0)
{
LogArrayValues(InteractionTarget.Values, "targetInteraction");
}
}
}
_counter++;
#endif
Console.WriteLine(_counter);
Distribution variantTarget = null;
if (!Model.Setting.NoVariantCalculation)
{
var localScaledVariants = new double[ScaledVariantTarget.Values.Length];
Array.Copy(ScaledVariantTarget.Values, localScaledVariants, ScaledVariantTarget.Values.Length);
variantTarget = new Distribution(localScaledVariants);
}
var doubleVal = new double[Model.Setting.NumberOfFeatures];
var interacVal = new double[Model.Setting.NumberOfInteractions];
var values = new XReal(solution);
for (var i = 0; i < Model.Setting.NumberOfFeatures; i++)
{
doubleVal[i] = values.GetValue(i);
}
for (var i = 0; i < Model.Setting.NumberOfInteractions; i++)
{
interacVal[i] = values.GetValue(i + Model.Setting.NumberOfFeatures);
}
//the distribution from the NSGA2
var nsgaFv = new Distribution(doubleVal);
var interacDist = new Distribution(interacVal);
var watch = Stopwatch.StartNew();
//calculate the variant values
Distribution variantResult = null;
if (_counter > 9500)
{
var variantValuesWithoutInteraction = FeatureMatrix.Dot(doubleVal);
if (Model.Setting.NumberOfInteractions > 0)
{
var interacVals = InteractionMatrix.Dot(interacVal);
var variantResults = variantValuesWithoutInteraction.Add(interacVals);
variantResult = new Distribution(variantResults);
}
else
{
variantResult = new Distribution(variantValuesWithoutInteraction);
}
}
watch.Stop();
calcTime += watch.ElapsedMilliseconds;
//scale the variant target distribution to the size of the calculated variant distribution
var scaleWatch = Stopwatch.StartNew();
FMScaling fms = new FMScaling(Model);
if (Model.Setting.ScaleToGlobalMinMax)
{
var change = false;
if (currentMin > variantResult.Values.Min())
{
currentMin = variantResult.Values.Min();
change = true;
}
if (currentMax < variantResult.Values.Max())
{
currentMax = variantResult.Values.Max();
change = true;
}
if (change)
{
ScaledVariantTarget = FMScaling.InteractionToScale(ScaledVariantTarget, currentMin, currentMax);
}
}
else
{
if (!Model.Setting.NoVariantCalculation)
{
variantTarget = FMScaling.InteractionToScale(variantTarget, variantResult.Values.Min(),
variantResult.Values.Max());
}
}
scaleWatch.Stop();
scaleTime += scaleWatch.ElapsedMilliseconds;
IntergenSolution s = (IntergenSolution)solution;
s.FoundAtEval = _counter;
var testWatch = Stopwatch.StartNew();
//calculate the fitness values for features and variants
var fc = new FitnessCalculator(Model, _counter);
var fitnessValues = fc.Calculate(variantResult, variantTarget, nsgaFv, FeatureTarget, interacDist, InteractionTarget);
solution.Objective[0] = fitnessValues.FeatureVal;
if (Model.Setting.NoVariantCalculation)
{
if (Model.Setting.NumberOfInteractions > 0)
{
solution.Objective[1] = fitnessValues.InteracVal;
}
}
else
{
//interacs and variants
if (Model.Setting.NumberOfInteractions > 0)
{
solution.Objective[1] = fitnessValues.InteracVal;
solution.Objective[2] = fitnessValues.VariantVal;
}
else
{
solution.Objective[1] = fitnessValues.VariantVal;
}
}
/* if (Model.Setting.UseKs)
* {
* PerfomKs(variantResult, variantTarget, nsgaFv, FeatureTarget, solution);
*
* }
* else if (Model.Setting.UseCmv)
* {
* PerformCmv(variantResult, variantTarget, nsgaFv, FeatureTarget, InteractionTarget, interacDist, solution);
* }
* else if (Model.Setting.UseEuclidean) {
* var bd = new BinnedDistance(variantTarget, variantResult, Model.VariantDynamicHist, _counter);
* var bd2 = new BinnedDistance(nsgaFv, FeatureTarget, Model.FeaturesDynamicHist, _counter);
*
* solution.Objective[0] = bd2.EuclidianDist();
* solution.Objective[1] = bd.EuclidianDist();
* if (Model.Setting.NumberOfInteractions > 0)
* {
* var interac = new BinnedDistance(interacDist, InteractionTarget, Model.InteracDynamicHist, _counter);
* solution.Objective[2] = interac.EuclidianDist();
* }
* }
* else if (Model.Setting.UseChiSquared)
* {
* var bd = new BinnedDistance(variantTarget, variantResult, Model.VariantDynamicHist,
* _counter);
* var bd2 = new BinnedDistance(nsgaFv, FeatureTarget, Model.FeaturesDynamicHist, _counter);
*
*
* solution.Objective[0] = bd2.ChiSquaredDist();
* solution.Objective[1] = bd.ChiSquaredDist();
* if (Model.Setting.NumberOfInteractions > 0)
* {
* var interac = new BinnedDistance(interacDist, InteractionTarget, Model.InteracDynamicHist, _counter);
* solution.Objective[2] = interac.ChiSquaredDist();
* }
* }
* else if (Model.Setting.EuclAndCmv)
* {
* PerformCmv(variantTarget, variantResult, nsgaFv, FeatureTarget, InteractionTarget, interacDist, solution);
*
* var variant = new BinnedDistance(variantTarget, variantResult, Model.VariantDynamicHist, _counter);
* var feat = new BinnedDistance(nsgaFv, FeatureTarget, Model.FeaturesDynamicHist, _counter);
* var interac = new BinnedDistance(interacDist, InteractionTarget, Model.InteracDynamicHist, _counter);
*
* solution.Objective[3] = feat.EuclidianDist();
* solution.Objective[4] = variant.EuclidianDist();
* solution.Objective[5] = interac.EuclidianDist();
* }
* else if (Model.Setting.ChiAndCmv)
* {
* PerformCmv(variantTarget, variantResult, nsgaFv, FeatureTarget, InteractionTarget, interacDist, solution);
* var variant = new BinnedDistance(variantTarget, variantResult, Model.VariantDynamicHist,_counter);
* var feat = new BinnedDistance(nsgaFv, FeatureTarget, Model.FeaturesDynamicHist, _counter);
* var interac = new BinnedDistance(interacDist, InteractionTarget, Model.InteracDynamicHist, _counter);
* solution.Objective[3] = feat.ChiSquaredDist();
* solution.Objective[4] = variant.ChiSquaredDist();
* solution.Objective[5] = interac.ChiSquaredDist();
* }
*/
testWatch.Stop();
fitnessTime += testWatch.ElapsedMilliseconds;
//report our progress to the model for GUI
if (_counter % 200 == 0)
{
if (!Model.Setting.Parallel /*|| (Model.Parallel && parallelIndex % 50 == 0) */)
{
if (Model.Setting.DrawDensity && Model.Setting.DrawHistogram)
{
RIntegrator.FeatureHistAndDens(nsgaFv.Values, FeatureTarget.Values);
RIntegrator.VariantHistAndDens(variantResult.Values, variantTarget.Values);
}
else if (Model.Setting.DrawDensity)
{
RIntegrator.PlotFeatureTarget(nsgaFv.Values, FeatureTarget.Values, Model.Setting.FeatureAdjust);
if (Model.Setting.NoVariantCalculation)
{
if (_counter > 9500)
{
RIntegrator.PlotVariantTarget(variantResult.Values);
}
}
else
{
RIntegrator.PlotVariantTarget(variantResult.Values, variantTarget.Values);
}
if (Model.Setting.NumberOfInteractions > 0)
{
RIntegrator.PlotInteracTarget(interacDist.Values, InteractionTarget.Values);
}
}
else if (Model.Setting.DrawHistogram)
{
RIntegrator.FeatureComparisonHist(nsgaFv.Values, FeatureTarget.Values);
if (Model.Setting.NoVariantCalculation)
{
RIntegrator.PlotVariantTarget(variantResult.Values);
}
else
{
RIntegrator.VariantComparisonHisto(variantResult.Values, variantTarget.Values);
}
}
else
{
}
//ReportProgress(solution);
}
ReportProgress(solution, fitnessValues);
//Worker.ReportProgress((int)(counter * 100 / (double)Model.MaxEvaluations), new UserProgress { VariantP = solution.Objective[1], FeatureP = solution.Objective[0] });
}
if (_counter == Model.Setting.MaxEvaluations)
{
Console.WriteLine("FitnessTime: " + fitnessTime);
Console.WriteLine("ScaleTime: " + scaleTime);
Console.WriteLine("CalcTime: " + calcTime);
}
var cont = new SolutionContainer
{
//TargetVariant = variantTarget,
Variant = variantResult,
Features = nsgaFv,
Interaction = interacDist,
//TargetFeatures = FeatureTarget,
FeatureTVal = fitnessValues.FeatureVal,
InteracTVal = fitnessValues.InteracVal,
VariantTVal = fitnessValues.VariantVal,
CalcTime = watch.ElapsedMilliseconds,
FitnessTime = testWatch.ElapsedMilliseconds,
ScaleTime = scaleWatch.ElapsedMilliseconds,
TestName = GetUsedTest(),
FoundAtEval = _counter,
};
//if (_counter > 7000)
// {
// cont.Write(Model.Setting.LogFolder + "evalstep" + _counter + ".json");
// }
//cont = null;
//Console.WriteLine(cont.FoundAtEval + "\t" + cont.FeatureTVal + "\t" + cont.VariantTVal);
if (Model.Setting.ChiAndCmv || Model.Setting.EuclAndCmv)
{
cont.AdditionalFeatureCmv = solution.Objective[4];
cont.AdditionalVariantCmv = solution.Objective[5];
}
//if (!Model.Setting.Parallel)Model.History.Add(counter, cont);
Model.AddSolutionToHistory(cont);
if (!Model.Setting.Logging)
{
return;
}
var usedTest = GetUsedTest();
//LogArrayValues(nsgaFV.Values, "Features");
//LogArrayValues(variantResult.Values, "Variants");
//LogArrayValues(ScaledVariantTarget.Values, "targetVariants");
if (Model.Setting.NumberOfInteractions > 0)
{
LogSingleValue(fitnessValues.InteracVal, "InteracFitn" + usedTest);
}
if (!Model.Setting.NoVariantCalculation)
{
LogSingleValue(fitnessValues.VariantVal, "VarFitn" + usedTest);
}
LogSingleValue(fitnessValues.FeatureVal, "FeatFitn" + usedTest);
}