private ANNTrainer haveChild(
ANNTrainer parent1,
ANNTrainer parent2,
double mutateChance,
double mutateMaxFactor
)
{
ANNTrainer child = new ANNTrainer(
new ANN(
parent1.nn.getLayers()
)
);
for (int i = 0; i < parent1.nn._biases.Count(); ++i)
{
child.nn._biases[i] = Rand.Chance(.5) ? parent1.nn._biases[i] : parent2.nn._biases[i];
}
for (int i = 0; i < parent1.nn._weights.Count(); ++i)
{
child.nn._weights[i] = Rand.Chance(.5) ? parent1.nn._weights[i] : parent2.nn._weights[i];
}
return(mutate(child, mutateChance, mutateMaxFactor));
}
public void Start()
{
if (gameOverPanel != null)
{
gameOverPanel.SetActive(false);
}
_zones = GameObject.FindGameObjectsWithTag(
"ScoreZone"
).ToList <GameObject>(
).ConvertAll <SafeZone>(
(go) => go.GetComponent <SafeZone>()
);
_buildingMask = 1 << LayerMask.NameToLayer("Building");
_bart = GameObject.Find("Bart");
_bartZone = _bart.GetComponent <Enemy>().zone;
_marge = GameObject.Find("Marge");
_margeZone = _marge.GetComponent <Enemy>().zone;
_moe = GameObject.Find("Moe");
_moeZone = _moe.GetComponent <Enemy>().zone;
_moeMovement = _moe.GetComponent <Movement>();
_burns = GameObject.Find("Burns");
_burnsZone = _burns.GetComponent <Enemy>().zone;
_burnsMovement = _burns.GetComponent <Movement>();
_homerFactory = GameObject.Find("HomerFactory").GetComponent <HomerFactory>();
const int populationCount = 8;
List <int> layers = new List <int>();
layers.Add(8); // Input layer
layers.Add(18); // Hidden layer
layers.Add(2); // Output layer
const double mutateChance = .4;
const double mutateMaxFactor = .4;
const double migrationPercentage = 0;
const double childrenPercentage = .19;
const int generations = 1000;
Action <ANNTrainer, Action <double> > train = (ANNTrainer t, Action <double> callback) => {
_nn = t;
_callback = callback;
_startGame();
};
new EO(
//populationCount,
//layers,
@"C:\Users\Public\EO\mc40mmf40mp0cp19g1000p8generation8.txt",
mutateChance,
mutateMaxFactor,
migrationPercentage,
childrenPercentage,
generations,
train
);
}
public void Start()
{
if (gameOverPanel != null) {
gameOverPanel.SetActive(false);
}
_zones = GameObject.FindGameObjectsWithTag(
"ScoreZone"
).ToList<GameObject>(
).ConvertAll<SafeZone>(
(go) => go.GetComponent<SafeZone>()
);
_buildingMask = 1 << LayerMask.NameToLayer("Building");
_bart = GameObject.Find("Bart");
_bartZone = _bart.GetComponent<Enemy>().zone;
_marge = GameObject.Find("Marge");
_margeZone = _marge.GetComponent<Enemy>().zone;
_moe = GameObject.Find("Moe");
_moeZone = _moe.GetComponent<Enemy>().zone;
_moeMovement = _moe.GetComponent<Movement>();
_burns = GameObject.Find("Burns");
_burnsZone = _burns.GetComponent<Enemy>().zone;
_burnsMovement = _burns.GetComponent<Movement>();
_homerFactory = GameObject.Find("HomerFactory").GetComponent<HomerFactory>();
const int populationCount = 8;
List<int> layers = new List<int>();
layers.Add(8); // Input layer
layers.Add(18); // Hidden layer
layers.Add(2); // Output layer
const double mutateChance = .4;
const double mutateMaxFactor = .4;
const double migrationPercentage = 0;
const double childrenPercentage = .19;
const int generations = 1000;
Action<ANNTrainer, Action<double>> train = (ANNTrainer t, Action<double> callback) => {
_nn = t;
_callback = callback;
_startGame();
};
new EO(
//populationCount,
//layers,
@"C:\Users\Public\EO\mc40mmf40mp0cp19g1000p8generation8.txt",
mutateChance,
mutateMaxFactor,
migrationPercentage,
childrenPercentage,
generations,
train
);
}
private ANNTrainer mutate(ANNTrainer trainer, double mutateChance, double mutateMaxFactor)
{
for (int i = 0; i < trainer.nn._biases.Count(); ++i)
{
trainer.nn._biases[i] = trainer.nn._biases[i].map(
(double b) => {
return(Rand.Chance(mutateChance) ? b + Rand.Range(mutateMaxFactor) : b);
}
);
}
for (int i = 0; i < trainer.nn._weights.Count(); ++i)
{
trainer.nn._weights[i] = trainer.nn._weights[i].map(
(double w) => {
return(Rand.Chance(mutateChance) ? w + Rand.Range(mutateMaxFactor) : w);
}
);
}
return(trainer);
}
private ANNTrainer haveChild(
ANNTrainer parent1,
ANNTrainer parent2,
double mutateChance,
double mutateMaxFactor
)
{
ANNTrainer child = new ANNTrainer(
new ANN(
parent1.nn.getLayers()
)
);
for (int i = 0; i < parent1.nn._biases.Count(); ++i) {
child.nn._biases[i] = Rand.Chance(.5) ? parent1.nn._biases[i] : parent2.nn._biases[i];
}
for (int i = 0; i < parent1.nn._weights.Count(); ++i) {
child.nn._weights[i] = Rand.Chance(.5) ? parent1.nn._weights[i] : parent2.nn._weights[i];
}
return mutate(child, mutateChance, mutateMaxFactor);
}
private void _initialize(
double mutateChance,
double mutateMaxFactor,
double migrationPercentage,
double childrenPercentage,
int generations,
Action<ANNTrainer, Action<double>> train,
List<ANN> population
)
{
string uniqueId = "mc" + ((int) (mutateChance * 100)) +
"mmf" + ((int) (mutateMaxFactor * 100)) +
"mp" + ((int) (migrationPercentage * 100)) +
"cp" + ((int) (childrenPercentage * 100)) +
"g" + generations + "p" + population.Count();
int numMigrate = Mathf.RoundToInt((float) migrationPercentage * population.Count());
int numChildren = Mathf.RoundToInt((float) childrenPercentage * population.Count());
System.IO.File.WriteAllText(
@"C:\Users\Public\EO\" + uniqueId + "generation0.txt",
String.Join(
",\n",
population.Select(
nn => nn.ToString()
).ToArray()
)
);
// This lambda soup is to allow the training to be asynchronous,
// I tried using C#'s await, yeild and Task, but was unable to get
// the desired behvaiour;
Action<int> trainGeneration;
trainGeneration = (int gen) => {
List<ANNTrainer> trained = new List<ANNTrainer>();
Action populationFinished = () => {
// Biggest to smalest
trained.Sort((a, b) => b.score.CompareTo(a.score));
ANNTrainer best = trained[0];
ANNTrainer second = trained[1];
// Add migrants
for (int i = 1; i <= numMigrate; ++i) {
Debug.Log("Adding migrant at " + (trained.Count() - i));
trained[trained.Count() - i] = new ANNTrainer(new ANN(best.nn.getLayers()));
}
for (int i = 1; i <= numChildren; ++i) {
Debug.Log("Adding child at " + (trained.Count() - i - numMigrate));
trained[trained.Count() - i - numMigrate] = this.haveChild(
best,
second,
mutateChance,
mutateMaxFactor
);
}
population = trained.ConvertAll<ANN>(trainer => trainer.nn);
System.IO.File.WriteAllText(
@"C:\Users\Public\EO\" + uniqueId + "generation" + gen + ".txt",
String.Join(
",\n",
population.Select(
nn => nn.ToString()
).ToArray()
)
);
if (gen + 1 < generations) {
trainGeneration(gen + 1);
}
};
Action<int> trainPopulation;
trainPopulation = (int pop) => {
Debug.Log("Training Generation " + gen + " Pop " + pop);
ANNTrainer trainer = new ANNTrainer(population[pop]);
train(
trainer,
(double score) => {
Debug.Log ("Pop " + pop + " scored " + score);
trainer.score = score;
trained.Add(trainer);
if (pop + 1 < population.Count()) {
trainPopulation(pop + 1);
} else {
populationFinished();
}
}
);
};
trainPopulation(0);
};
trainGeneration(0);
}
private ANNTrainer mutate(ANNTrainer trainer, double mutateChance, double mutateMaxFactor)
{
for (int i = 0; i < trainer.nn._biases.Count(); ++i) {
trainer.nn._biases[i] = trainer.nn._biases[i].map(
(double b) => {
return Rand.Chance(mutateChance) ? b + Rand.Range(mutateMaxFactor) : b;
}
);
}
for (int i = 0; i < trainer.nn._weights.Count(); ++i) {
trainer.nn._weights[i] = trainer.nn._weights[i].map(
(double w) => {
return Rand.Chance(mutateChance) ? w + Rand.Range(mutateMaxFactor) : w;
}
);
}
return trainer;
}
private void _initialize(
double mutateChance,
double mutateMaxFactor,
double migrationPercentage,
double childrenPercentage,
int generations,
Action <ANNTrainer, Action <double> > train,
List <ANN> population
)
{
string uniqueId = "mc" + ((int)(mutateChance * 100)) +
"mmf" + ((int)(mutateMaxFactor * 100)) +
"mp" + ((int)(migrationPercentage * 100)) +
"cp" + ((int)(childrenPercentage * 100)) +
"g" + generations + "p" + population.Count();
int numMigrate = Mathf.RoundToInt((float)migrationPercentage * population.Count());
int numChildren = Mathf.RoundToInt((float)childrenPercentage * population.Count());
System.IO.File.WriteAllText(
@"C:\Users\Public\EO\" + uniqueId + "generation0.txt",
String.Join(
",\n",
population.Select(
nn => nn.ToString()
).ToArray()
)
);
// This lambda soup is to allow the training to be asynchronous,
// I tried using C#'s await, yeild and Task, but was unable to get
// the desired behvaiour;
Action <int> trainGeneration;
trainGeneration = (int gen) => {
List <ANNTrainer> trained = new List <ANNTrainer>();
Action populationFinished = () => {
// Biggest to smalest
trained.Sort((a, b) => b.score.CompareTo(a.score));
ANNTrainer best = trained[0];
ANNTrainer second = trained[1];
// Add migrants
for (int i = 1; i <= numMigrate; ++i)
{
Debug.Log("Adding migrant at " + (trained.Count() - i));
trained[trained.Count() - i] = new ANNTrainer(new ANN(best.nn.getLayers()));
}
for (int i = 1; i <= numChildren; ++i)
{
Debug.Log("Adding child at " + (trained.Count() - i - numMigrate));
trained[trained.Count() - i - numMigrate] = this.haveChild(
best,
second,
mutateChance,
mutateMaxFactor
);
}
population = trained.ConvertAll <ANN>(trainer => trainer.nn);
System.IO.File.WriteAllText(
@"C:\Users\Public\EO\" + uniqueId + "generation" + gen + ".txt",
String.Join(
",\n",
population.Select(
nn => nn.ToString()
).ToArray()
)
);
if (gen + 1 < generations)
{
trainGeneration(gen + 1);
}
};
Action <int> trainPopulation;
trainPopulation = (int pop) => {
Debug.Log("Training Generation " + gen + " Pop " + pop);
ANNTrainer trainer = new ANNTrainer(population[pop]);
train(
trainer,
(double score) => {
Debug.Log("Pop " + pop + " scored " + score);
trainer.score = score;
trained.Add(trainer);
if (pop + 1 < population.Count())
{
trainPopulation(pop + 1);
}
else
{
populationFinished();
}
}
);
};
trainPopulation(0);
};
trainGeneration(0);
}
