private void RecalculateFitness(Representation reference)
{
// Recalculates fitness of members based on similarity criteria
double weight = 0.4;
foreach (Representation member in currentPopulation)
{
int distance = StaticOperations.TotalDistanceBetweenPaths(reference, member);
// The higher the distance, the better
double newFitness = member.Fitness * (weight + (1 - weight) * distance);
member.CombinedFitness = newFitness;
}
}
public List <Representation> DuplicationElimination(int mutationAttempts,
int similarMemberCountThreshold, int similarityThreshold)
{
// New population will consist of
// a) elite ten percent of unique old members
// b) similar descendants will be repeatadly mutated to achieve higher chance of
// keeping multiple candidates
// Choose ten percent of unique members
List <Representation> eliteMembers = UniqueEliteSelection(oldGeneration.Count / 10 < 6 ? 6 : oldGeneration.Count / 10);
// Test all descendants
List <Representation> survivingDescendants = new List <Representation>();
Decoder decoder = new Decoder();
Evaluation evaluation = new Evaluation();
for (int i = 0; i < descendants.Count; i++)
{
Representation descendant = descendants[i];
double startFitness = descendant.Fitness;
List <double> distances = new List <double>();
foreach (Representation oldMember in oldGeneration)
{
distances.Add(StaticOperations.TotalDistanceBetweenPaths(descendant, oldMember));
}
int similarCount = distances.Count(item => item <= similarityThreshold);
if (similarCount > similarMemberCountThreshold)
{
GeneticOperator geneticOperator = new GeneticOperator();
int attempts = 0;
while (similarCount > similarMemberCountThreshold && attempts < mutationAttempts)
{
descendant = geneticOperator.MutateMember(descendant);
distances.Clear();
foreach (Representation oldMember in oldGeneration)
{
distances.Add(StaticOperations.TotalDistanceBetweenPaths(descendant, oldMember));
}
similarCount = distances.Count(item => item <= similarityThreshold);
attempts++;
// Prevents repetitive mutation to have an aspiring solution stuck at lower fitness values
if (IsUnique(descendant) && evaluation.EvaluatePath(decoder.DecodeRepresentation(descendant)) > startFitness)
{
break;
}
}
survivingDescendants.Add(descendant);
}
else
{
survivingDescendants.Add(descendant);
}
}
// Some members were mutated and their fitness is unknown
foreach (Representation descendant in survivingDescendants)
{
if (descendant.Fitness == 0)
{
descendant.Fitness = evaluation.EvaluatePath(decoder.DecodeRepresentation(descendant));
}
}
survivingDescendants.OrderByDescending(item => item.Fitness);
eliteMembers.AddRange(survivingDescendants.Take(nextGenerationCount - eliteMembers.Count));
return(eliteMembers);
}
public List <Representation> NextGeneration()
{
// New generation will consist of both old and new members
// Genetic operators generate as many descendants as there are parents
// Every descendant will compete against a small selection of old generation
// From that selection, the most similar member is chosen and competes against the descendant
// Winner remains in population
List <Representation> newGeneration = StaticOperations.CloneList(oldGeneration);
List <Representation> eliteMembers = UniqueEliteSelection(oldGeneration.Count / 10 < 6 ? 6 : oldGeneration.Count / 10);
foreach (Representation representation in eliteMembers)
{
Representation member = newGeneration.Find(item => item.Values.SequenceEqual(representation.Values));
newGeneration.Remove(member);
}
List <Representation> survivingDescendatns = new List <Representation>();
descendants.Shuffle();
descendants = descendants.Skip(eliteMembers.Count).ToList();
Random random = new Random();
foreach (Representation descendant in descendants)
{
// Generate random selection of old members
int sampleCount = random.Next(3, 7);
int[] sampleIndices = new int[sampleCount];
for (int i = 0; i < sampleCount; i++)
{
sampleIndices[i] = random.Next(0, newGeneration.Count);
}
List <Representation> sampleMembers = new List <Representation>();
for (int i = 0; i < sampleCount; i++)
{
sampleMembers.Add(newGeneration[sampleIndices[i]]);
}
// Calculate, which of the selected members is more similar to current member
int[] distancesField = new int[sampleCount];
for (int i = 0; i < sampleMembers.Count; i++)
{
distancesField[i] = StaticOperations.TotalDistanceBetweenPaths(descendant, sampleMembers[i]);
}
int minDistanceIndex = 0;
for (int i = 0; i < distancesField.Length; i++)
{
if (distancesField[minDistanceIndex] < distancesField[i])
{
minDistanceIndex = i;
}
}
// Stochastic competition
// Descendant and old member fight in stochastic probabilistic tournament
// based on their fitness
int selectingNumber = random.Next(0, (int)(descendant.Fitness + sampleMembers[minDistanceIndex].Fitness));
if (selectingNumber <= (int)descendant.Fitness)
{
survivingDescendatns.Add(descendant);
newGeneration.Remove(sampleMembers[minDistanceIndex]);
}
// Compare fitness of descendant and calculated similar member
// Winner goes into new generation
//if (descendant.Fitness > sampleMembers[minDistanceIndex].Fitness)
//{
// survivingDescendatns.Add(descendant);
// newGeneration.Remove(sampleMembers[minDistanceIndex]);
//}
}
// New generation consists of old generation members that won their duels and
// descendants that eliminated losing old generation members
newGeneration.AddRange(survivingDescendatns);
newGeneration.AddRange(eliteMembers);
return(newGeneration);
}
