/// <summary>
/// Performs crossover to generate a new child. The crossover type is specifed in the
/// Settings property.
/// </summary>
/// <param name="mate">The individual to crossover with.</param>
/// <returns>A new child individual</returns>
public IIndividual Crossover(IIndividual mate)
{
if (mate.GetType() != typeof(LanderIndividual))
{
throw new ArgumentException("Mate must be a LanderIndividual");
}
// Get the weights in a easier to use form.
this.weights = this.neuralNet.GetAllWeights();
List<double> mateWeights = ((LanderIndividual)mate).neuralNet.GetAllWeights();
// Create the child weights with a copy of this individual's weights initially
List<double> childWeights = new List<double>(this.weights);
switch (this.settings.CrossoverAlgorithm)
{
case LanderIndividualSettings.CrossoverType.OnePoint:
// Copy the mate's weights into the first section
int crossoverPoint = this.RandomGenerator.Next(this.weights.Count);
for (var i = 0; i < crossoverPoint; i++)
{
childWeights[i] = mateWeights[i];
}
break;
case LanderIndividualSettings.CrossoverType.TwoPoint:
// Copy the mate's weights into the first section
int firstPoint = this.RandomGenerator.Next(this.weights.Count);
int secondPoint = this.RandomGenerator.Next(this.weights.Count);
if (secondPoint < firstPoint)
{
int temp = firstPoint;
firstPoint = secondPoint;
secondPoint = temp;
}
for (var i = 0; i < firstPoint; i++)
{
childWeights[i] = mateWeights[i];
}
for (var i = secondPoint; i < childWeights.Count; i++)
{
childWeights[i] = mateWeights[i];
}
break;
case LanderIndividualSettings.CrossoverType.Uniform:
for (int i = 0; i < childWeights.Count; i++)
{
if (RandomGenerator.NextDouble() < 0.5)
{
childWeights[i] = mateWeights[i];
}
}
break;
default:
throw new ArgumentException("Lander individual crossover type not supported");
}
// Create the new child
LanderIndividual child = new LanderIndividual(this.RandomGenerator);
child.Settings = this.Settings;
child.neuralNet.SetAllWeights(childWeights);
return child;
}
/// <summary>
/// Returns a deep clone of the individual.
/// </summary>
/// <returns>A deep clone of the individual.</returns>
public object Clone()
{
LanderIndividual clone = new LanderIndividual(this.RandomGenerator);
clone.neuralNet = new NeuralNetwork();
clone.settings = this.settings;
// NOTE: network topology defined here
this.weights = this.neuralNet.GetAllWeights();
clone.neuralNet.InputCount = this.neuralNet.InputCount;
clone.neuralNet.OutputCount = this.neuralNet.OutputCount;
clone.neuralNet.AddHiddenLayer(5);
clone.neuralNet.SetAllWeights(this.weights);
return clone;
}