public override IIndividual Mutate(float probability, float sigma, Random r)
{
var newInd = new AdaptiveCartesianIndividual(this);
//Mutate mutation (add a number between (-0.1,0.1) )
//F****d up side effect of not setting the probability to the child but the parent instead.
// The children had worse values as sigmas skyroketted, keeping the parents alive even longer
// The parents were making they're children worse off so they could keep surviving :/
newInd.MutationProbability = (MutationProbability + (r.NextDouble() - 0.5) / 5).Clamp(0, 1);
newInd.AverageCrossOverProbabilty = (AverageCrossOverProbabilty + (r.NextDouble() - 0.5) / 5).Clamp(0, 1);
//Mutate sigmas based on Tau
for (int i = 0; i < Sigmas.Length; ++i)
{
if (r.ProbabilityPass((float)newInd.MutationProbability))
{
double modFactor = r.GausianNoise(Tau) + 1;
newInd.Sigmas[i] = newInd.Sigmas[i] * modFactor;
}
}
//Mutate actual values based on Sigmas
for (int i = 0; i < Values.Length; ++i)
{
if (r.ProbabilityPass((float)newInd.MutationProbability))
{
//Here we were using the parent's sigmas...
newInd.Values[i] += r.GausianNoise(newInd.Sigmas[i]);
}
}
return(newInd);
}
public AdaptiveCartesianIndividual(AdaptiveCartesianIndividual other) : base(other)
{
Sigmas = other.Sigmas.ToArray();
MutationProbability = 0.6;
AverageCrossOverProbabilty = 0.66;
}
public static IIndividual CrossOver(List <IIndividual> parents, Random r)
{
var typedParents = parents.OfType <AdaptiveCartesianIndividual>().ToList();
int newLength =
typedParents
.PickRandom(r)
.Values
.Length;
float aveCrossOverProb = (float)typedParents.Select(i => i.AverageCrossOverProbabilty).Average();
var newInd = new AdaptiveCartesianIndividual(newLength, r: r, empty: true);
bool takeAverage = false;
//Go through the values and the sigma
for (var ii = 0; ii < newLength; ++ii)
{
takeAverage = r.ProbabilityPass(aveCrossOverProb);
if (takeAverage)
{
//Take average of parents respective alleles
double valueTot = 0.0;
double sigmaTot = 0.0;
for (var kk = 0; kk < parents.Count; ++kk)
{
AdaptiveCartesianIndividual par = typedParents[kk];
//TODO: is the absence of value zero?
if (ii < par.Values.Length)
{
valueTot += par.Values[ii];
sigmaTot += par.Sigmas[ii];
}
}
//TODO: is the absence of value zero?
newInd.Values[ii] = valueTot / parents.Count;
newInd.Sigmas[ii] = sigmaTot / parents.Count;
}
else //Take single parent gene randomly
{
AdaptiveCartesianIndividual par = typedParents[r.Next(0, typedParents.Count)];
//TODO: Is the absence of value zero?
newInd.Values[ii] = ii < par.Values.Length
? par.Values[ii]
: 0;
newInd.Sigmas[ii] = ii < par.Sigmas.Length
? par.Sigmas[ii]
: DefaultSigma;
}
}
if (r.ProbabilityPass(aveCrossOverProb))
{
newInd.MutationProbability = typedParents.Select(i => i.MutationProbability).Average();
}
else
{
newInd.MutationProbability = typedParents.PickRandom(r).MutationProbability;
}
if (r.ProbabilityPass(aveCrossOverProb))
{
newInd.AverageCrossOverProbabilty = typedParents.Select(i => i.AverageCrossOverProbabilty).Average();
}
else
{
newInd.AverageCrossOverProbabilty = typedParents.PickRandom(r).AverageCrossOverProbabilty;
}
return(newInd);
}
