public void OnRoundStart(Dictionary <int, FighterController> fighters)
{
foreach (KeyValuePair <int, FighterController> entry in fighters)
{
if (!(entry.Value is NeuralFighterController))
{
continue;
}
initializePopulation(entry.Key);
NeuralFighterController nfc = (NeuralFighterController)entry.Value;
nfc.NeuralNetwork = GetLastFittestNetwork(entry.Key);
// First update the previous elite, then we can run another generation.
string fighterIdentifier = fighters[nfc.GetTarget(fighters)].Identifier;
float[] weights = fighterPopulations[entry.Key].FittestGenome.Weights;
addPreviousElite(fighterIdentifier, weights);
GameManager.Instance.StartCoroutine(DoGeneration(entry.Key, fighters[nfc.GetTarget(fighters)]));
}
}
public IEnumerator DoGeneration(int identifier, FighterController opponentFighterController)
{
int simulationsPerFrame = 5;
int simulationsLeft = 5;
NeuralNetworkPopulation nnp = fighterPopulations[identifier];
for (int i = 0; i < nnp.Population.Count; i++)
{
int turn = 0;
NeuralFighterController fighter = GameManager.Instance.gameObject.AddComponent <NeuralFighterController>();
FighterController opponentFighter = (FighterController)GameManager.Instance.gameObject.AddComponent(opponentFighterController.GetType());
Dictionary <int, FighterController> fighters = new Dictionary <int, FighterController> {
{ 0, fighter }, { 1, opponentFighter }
};
Dictionary <int, BattleAction> previousFigherActions = new Dictionary <int, BattleAction> {
{ 0, new BattleAction(ActionType.NOTHING, 1) }, { 1, new BattleAction(ActionType.NOTHING, 0) }
};
fighter.NeuralNetwork = nnp.Population[i];
fighter.Awake();
fighter.Start();
opponentFighter.Awake();
opponentFighter.Start();
while (fighter.Health > 0 && opponentFighter.Health > 0 && turn < NeuralParameters.MAX_TURNS)
{
BattleAction fighterAction = fighter.GetAction(fighters, previousFigherActions);
BattleAction opponentAction = opponentFighter.GetAction(fighters, previousFigherActions);
previousFigherActions[0] = fighterAction;
previousFigherActions[1] = opponentAction;
fighter.Health -= getReceivedDamage(fighterAction, opponentAction);
opponentFighter.Health -= getReceivedDamage(opponentAction, fighterAction);
turn++;
}
if (opponentFighter.Health < 0)
{
opponentFighter.Health = 0;
}
if (fighter.Health < 0)
{
fighter.Health = 0;
}
float fitness = fighter.Health - opponentFighter.Health - turn;
if (opponentFighter.Health > 0 && fighter.Health > 0)
{
if (fitness > 0)
{
fitness *= NeuralParameters.POSITIVE_OUT_OF_TIME_MULTIPLIER;
}
else
{
fitness *= NeuralParameters.NEGATIVE_OUT_OF_TIME_MULTIPLIER;
}
}
else if (opponentFighter.Health > 0 && fighter.Health <= 0)
{
fitness *= NeuralParameters.LOSE_MULTIPLIER;
}
nnp.Population[i].Fitness = fitness;
Genome genome = nnp.GeneticAlgorithm.Population[i];
genome.Fitness = fitness;
nnp.GeneticAlgorithm.Population[i] = genome;
Destroy(fighter);
Destroy(opponentFighter);
if (--simulationsLeft < 0)
{
yield return(null);
simulationsLeft = simulationsPerFrame;
}
}
nnp.Epoch();
nnp.EliteEpoch(previousElite);
Debug.Log(string.Format("Best fitness: {0} - Average fitness: {1} - Worst fitness: {2}", nnp.BestFitness, Mathf.Round(nnp.AverageFitness * 100f) / 100f, nnp.WorstFitness));
if (GameManager.Instance.DebugRun)
{
Debug.Break();
}
}
