//When a car collides kill the child
public void OnCollisionEnter(Collision col)
{
CalculateFitness();
ResetCarPosition();
//display the current child fitness
Display += "Child " + (currentChild + 1) + ": " + Fitness[currentChild].ToString("f2") + "\n";
PopulationText.GetComponent <Text>().text = Display;
//if all children have died
if (currentChild == Fitness.Length - 1)
{
int[] FitestParents = new int[2];
//find 2 of the fittest parents
FitestParents = FindFittestParents(Fitness);
GeneticNetwork Father = Cars[FitestParents[0]];
GeneticNetwork Mother = Cars[FitestParents[1]];
//create a new generation of children based on 2 fittest parents
for (int i = 0; i < Population; i++)
{
Fitness[i] = 0;
Cars[i] = new GeneticNetwork(Father, Mother, Probability);
}
//increment generation, reset current child and fitness display
Generation++;
currentChild = -1;
Display = "";
}
currentChild++;
}
//Initialise objects
void Start()
{
Rigidbody = GetComponent <Rigidbody>();
Outputs = new List <double>();
Car = new GeneticNetwork();
saveLoad = new GeneticWeights();
controller = new CarController();
Rigidbody.GetComponent <CarPhysics>().driver = Driver.AI;
position = new ResetPosition();
LoadChild();
}
public void Random2()
{
// ARRANGE
GeneticNetwork genetic = new GeneticNetwork();
// ACT
double random = genetic.GetRandomWeight();
// ASSERT
Assert.That(random, Is.InRange(-1f, 1f));
}
public void Sigmoid1()
{
// ARRANGE
GeneticNetwork genetic = new GeneticNetwork();
List <double> sums = new List <double> {
1.0,
6.3,
2.2,
0.00005
};
// ACT
List <double> outputs = genetic.Sigmoid(sums);
// ASSERT
foreach (var output in outputs)
{
Assert.That(output, Is.InRange(0f, 1f));
}
}
public void FeedForward2()
{
// ARRANGE
GeneticNetwork genetic = new GeneticNetwork();
double[] inputs = { 0.6, 0.4, 0.2, 0, 0 };
int expectedlength = 2;
// ACT
List <double> outputs = genetic.FeedForward(inputs);
// ASSERT
Assert.That(outputs.Count, Is.EqualTo(expectedlength));
foreach (var output in outputs)
{
Assert.That(output, Is.InRange(0f, 1f));
}
}
//Crossover & mutation of weights based on selected probability at end of generation
public GeneticNetwork(GeneticNetwork Father, GeneticNetwork Mother, float probability)
{
System.Random randomBool = new System.Random();
this.weights = new List <double[][]>();
//For each layer create new matrix
for (int i = 0; i < lengthLayers - 1; i++)
{
double[][] layerWeights = new double[layers[i]][];
//for each row
for (int j = 0; j < layers[i]; j++)
{
layerWeights[j] = new double[layers[i + 1]];
//and for each column
for (int k = 0; k < layers[i + 1]; k++)
{
//randomly crossover either mother or father
if (randomBool.Next(2) == 0)
{
layerWeights[j][k] = Father.weights[i][j][k];
}
else
{
layerWeights[j][k] = Mother.weights[i][j][k];
}
//mutate that layer to a random weight if it falls under the probability
if (Random.Range(0f, 1f) < probability)
{
layerWeights[j][k] = GetRandomWeight();
}
}
}
//Add that new weight to the new list of weights (children)
weights.Add(layerWeights);
}
}
public void Init1()
{
// ARRANGE
int population = 10;
GeneticNetwork[] genetic = new GeneticNetwork[population];
int expectedlength = 10;
// ACT
for (int i = 0; i < population; i++)
{
genetic[i] = new GeneticNetwork();
}
// ASSERT
Assert.That(genetic.Length, Is.EqualTo(expectedlength));
for (int i = 0; i < population; i++)
{
Assert.That(genetic[i].GetWeights().GetType() == typeof(List <double[][]>));
}
}
//Initialise various methods and arrays
void Start()
{
fitnessMeasure = LearningModeScript.fitnessMeasure;
Population = LearningModeScript.Population;
Probability = LearningModeScript.Probability;
nextChild.onClick.AddListener(NextChild);
Rigidbody = GetComponent <Rigidbody>();
controller = new CarController();
saveLoad = new GeneticWeights();
Rigidbody.GetComponent <CarPhysics>().driver = Driver.AI;
Fitness = new double[Population];
Outputs = new List <double>();
Position = transform.position;
//Create cars of size population
Cars = new GeneticNetwork[Population];
//Initiate list of weights of size pupulation
for (int i = 0; i < Population; i++)
{
Cars[i] = new GeneticNetwork();
}
}