protected override void UpdateQValue(State s, int ai, double newValue)
{
double reward = EnvironmentModeler.GetReward(base.m_prevState, base.m_curState,
SoccerAction.GetActionTypeFromIndex(ai, Params.MoveKings));
double dummy;
int prevStateGreedyAct = GetMaxQ(base.m_prevState, out dummy);
if (prevStateGreedyAct == ai) // add reward to fitness only if it was a greedy action
{
m_genomeSumRewards += reward;
}
if (this.m_client.Cycle % NeatExpParams.CyclesPerGenome == 0)
{
Console.Write("Gener: {0}, ", m_evoAlg != null ? m_evoAlg.Generation : 0);
int curGoalDiff = m_client.OurScore - m_client.OppScore;
this.GoalDiffGainedByNetworkToEvaluate = curGoalDiff - m_prevGoalDiff;
this.FitnessNetworkToEvaluate = m_genomeSumRewards;
m_prevGoalDiff = curGoalDiff;
m_eventNetFitnessReady.Set();
m_genomeSumRewards = 0.0;
m_eventNewNetReady.WaitOne();
}
}
public override void UpdateQ_QLearning(int prevActIndex)
{
double reward = EnvironmentModeler.GetReward(m_prevState, m_curState,
SoccerAction.GetActionTypeFromIndex(prevActIndex, Params.MoveKings));
int ai_star_other = EnvironmentModeler.GuessOpponentAction(m_prevOriginalState, m_curOriginalState, m_oppToMonitor);
UpdateIM_Learning(m_prevState, ai_star_other);
int ai_prime_other = GetMaxIMIndex(m_prevState);
double oldQ = GetQValue(m_prevState, prevActIndex, ai_star_other);
double bestQ = GetBestQ(m_curState, ai_prime_other);
double newQ = oldQ + Params.Alpha * (reward + Params.Gamma * bestQ - oldQ);
UpdateQValue(m_prevState, prevActIndex, ai_star_other, newQ);
}
public void UpdateQ_QLearning(State prevState, State curState, SoccerAction prevAct)
{
double reward = EnvironmentModeler.GetReward(prevState, curState, prevAct.ActionType);
this.UpdateQ_QLearning(reward, prevState, curState, prevAct);
}