private void HandleGoodEnough(ref bool goodEnough, PlayoutResult newFitness, GenerationMember member,
ref int goodCount)
{
if (Constants.SaveGoodOnes && !goodEnough && newFitness.SimpleFitness() > 0.93)
{
goodCount++;
if (goodCount > 50)
{
goodEnough = true;
}
Console.WriteLine($"Found extra good {newFitness.SimpleFitness()}");
SaveDna(goodCount, member.dna);
}
}
public float CombinedFitness(DNA initialDna)
{
float fitA = 1 - result.HpPercentage;
float fitB = (float)PlayoutResult.LengthSample(result.TotalTurns);
float fitness = (fitA + fitB) / 2;
if (!result.AllPlayed)
{
fitness = 0.0001f;
}
var low = DenseVector.Build.Dense(dna.Data.Count, 0);
var high = DenseVector.Build.Dense(dna.Data.Count, 1);
double maxDistance = Distance.Euclidean(low, high);
double distance = Distance.Euclidean(dna.Data, initialDna.Data);
double relativeDistance = distance / maxDistance;
double distanceFitness = 1 / (1 + Math.Exp(-20 * relativeDistance + 4));
return((fitness + (float)distanceFitness) / 2);
}
public GenerationMember(DNA dna, PlayoutResult result)
{
this.dna = dna;
this.result = result;
}
public void RunSimulatedAnnealing()
{
var copy = _initialDna.Clone();
copy.Randomize();
var current = new GenerationMember {
dna = copy,
result = CalculateFitness(_game, _initialDna, copy)
};
float T = Constants.InitialT;
List <double> plotT = new List <double>();
List <double> plotFit = new List <double>();
List <double> plotHpPercentage = new List <double>();
List <double> plotLength = new List <double>();
int goodCount = 0;
bool goodEnough = false;
var gameCopies = Enumerable.Range(0, Constants.TeamsPerGeneration)
.AsParallel()
.Select(_ => _game.DeepCopy())
.ToList();
var initialDnaCopies = Enumerable.Range(0, Constants.TeamsPerGeneration)
.Select(_ => _initialDna.Clone())
.ToList();
for (int i = 0; i < Constants.NumGenerations; i++)
{
var tpercentage = Math.Max(0, 1.0f - (float)i / Constants.NumGenerations);
T = Constants.InitialT * tpercentage;
var genWatch = new Stopwatch();
genWatch.Start();
var teamWatch = new Stopwatch();
teamWatch.Start();
if (Constants.RestartFailures && current.result.SimpleFitness() < Constants.FitnessThreshold)
{
current.dna.Randomize();
_restartCount++;
}
//_initialDna = Mutate(_initialDna, T);
var tmp = T;
var current1 = current;
var generation = Enumerable.Range(0, Constants.TeamsPerGeneration)
.AsParallel()
.Select(j => {
var newDna = Mutate(current1.dna);
var newFitness =
CalculateFitness(gameCopies[j], initialDnaCopies[j], newDna);
return(new GenerationMember(newDna, newFitness));
})
.ToList();
var newMax = PickBestMember(_game, _initialDna, generation);
HandleGoodEnough(ref goodEnough, newMax.result, current, ref goodCount);
float e = current.result.SimpleFitness();
float ep = newMax.result.SimpleFitness();
double probability = 1;
float delta = ep - e;
if (delta > 0)
{
probability = 1;
current = newMax;
}
else
{
probability = Math.Exp(-Math.Abs(delta) / T);
if (Constants.HillClimbing ^ Probability.Uniform(probability))
{
newMax.failCount = current.failCount;
current = newMax;
}
current.failCount++;
}
plotT.Add(T);
plotFit.Add(current.result.SimpleFitness());
plotHpPercentage.Add(1 - current.result.HpPercentage);
//plotLength.Add(current.result.TotalTurns);
plotLength.Add(PlayoutResult.LengthSample(current.result.TotalTurns));
PrintEvaluationResult(i, e, ep, probability, T, current);
}
Console.WriteLine($"Restarts: {_restartCount}");
if (Constants.GnuPlot)
{
var gnuplotConfigString = $"title '{Constants.NumGenerations} generations," +
$"T_s = {Constants.InitialT}'";
GnuPlot.HoldOn();
GnuPlot.Set($"xrange [{Constants.InitialT}:{T}] reverse",
$"title '{Constants.NumGenerations} generations, T_s = {Constants.InitialT}",
"key tmargin center horizontal");
GnuPlot.Plot(plotT.ToArray(), plotFit.ToArray(), $"title 'Fitness {Constants.NumGenerations}gen'");
GnuPlot.Plot(plotT.ToArray(), plotHpPercentage.ToArray(), $"title 'HP percentage'");
GnuPlot.Plot(plotT.ToArray(), plotLength.ToArray(), "title 'Game length'");
Console.ReadKey();
}
}
