public override int MakeAction(int numChoices)
{
// SetupPrivateGame sets "privategame" equal to actualgame.clonesecret(idx) and
// sets "privateiterator" equal to actualgameiterator.clone()
(CardGame privategame, GameIterator privateiterator) = perspective.GetPrivateGame();
int idx = perspective.GetIdx();
Debug.WriteLine("PIPMC making choice. items: " + numChoices);
double[][] inverseRankSum = new double[perspective.NumberOfPlayers()][];
for (int i = 0; i < perspective.NumberOfPlayers(); i++)
{
inverseRankSum[i] = new double[numChoices];
}
Debug.WriteLine("Start Monte");
// can parallellize here TODO ?
// FOR EACH POSSIBLE MOVE
for (int move = 0; move < numChoices; ++move)
{
Debug.WriteLine("iterating over item: " + move);
Parallel.For(0, NUMTESTS, i => //number of tests for certain decision
{
Debug.WriteLine("****Made Switch**** : " + i);
// JUST USING ONE CLONE SECRETGAME, CLONED FOR EACH MOVE
CardGame cg = privategame.Clone();
GameIterator cloneContext = privateiterator.Clone(cg);
// Make the chosen move
List <GameActionCollection> allOptions = cloneContext.BuildOptions();
allOptions[move].ExecuteAll();
cloneContext.PopCurrentNode();
// Assign the AI players for rollout game
for (int j = 0; j < numPlayers; j++)
{
cg.players[j].decision = new RandomPlayer(perspective);
}
Debug.WriteLine("Playing a simulated game");
while (!cloneContext.AdvanceToChoice())
{
cloneContext.ProcessChoice();
}
Debug.WriteLine("Simulated Game is Over");
// ProcessScore returns a sorted list
// where the winner is rank 0 for either min/max games.
var winners = cloneContext.ProcessScore();
Debug.WriteLine("past ProcessScore");
int topRank = 0;
lock (this)
{
for (int j = 0; j < numPlayers; ++j)
{
if (j != 0 && winners[j].Item1 != winners[j - 1].Item1)
{
topRank = j;
}
inverseRankSum[winners[j].Item2][move] += (((double)1) / (topRank + 1)) / NUMTESTS;
}
}
});
}
// FIND BEST (and worst) MOVE TO MAKE
var tup = MinMaxIdx(inverseRankSum[perspective.GetIdx()]);
// Record info for heuristic evaluation
RecordHeuristics(inverseRankSum);
return(tup.Item2);
}
public void RunSimulation()
{
// Each turn, need to check to see if we have enough information to make move using UCB
// If we do (movelist.count() == choicenum), and we check the stats of each move
// A predictable player is set for the currentplayers idx which wil chose the move determined by
// Movelist should be tuple array with each entry a state and a who played it
// Its key should be a state and the idx of the player in charge
HashSet <Tuple <CardGame, int> > visitedstates = new HashSet <Tuple <CardGame, int> >();
CardGame cg = privategame.Clone();
GameIterator gameIterator = privateiterator.Clone(cg);
for (int j = 0; j < numPlayers; j++)
{
cg.players[j].decision = new RandomPlayer(perspective);
}
int idxme = cg.currentPlayer.Peek().idx;
bool expand = true;
bool first = true;
// "Playing a simulated game"
while (!gameIterator.AdvanceToChoice())
{
int idx = cg.currentPlayer.Peek().idx;
if (idxme == idx)
{
List <GameActionCollection> allOptions = gameIterator.BuildOptions();
Tuple <CardGame, int>[] movelist = null;
int c = 0;
if (expand)
{
int choicenum = allOptions.Count;
Tuple <CardGame, int> deliberator = Tuple.Create <CardGame, int>(cg.Clone(), idx);
if (!movestatetree.Keys.Contains(deliberator))
{
movestatetree[deliberator] = new Tuple <CardGame, int> [choicenum];
}
movelist = movestatetree[deliberator];
//Console.WriteLine("Choice num: " + choicenum + " Movelist Count: " + movelist.Count(s => s != null));
if (movelist.Count(s => s != null) == choicenum)
{
// USE UCB
double bestscore = 0;
c = 0;
double totalplays = 0;
foreach (Tuple <CardGame, int> stateandplay in (movelist))
{
totalplays += plays[stateandplay];
}
totalplays = Math.Log(totalplays);
for (int i = 0; i < movelist.Length; i++)
{
Tuple <CardGame, int> stateandplay = movelist[i];
double temp = wins[stateandplay] / plays[stateandplay];
temp += Math.Sqrt(2 * totalplays / plays[stateandplay]);
if (temp > bestscore)
{
bestscore = temp;
c = i;
}
}
allOptions[c].ExecuteAll();
gameIterator.PopCurrentNode();
}
else
{
c = gameIterator.ProcessChoice();
}
}
else
{
c = gameIterator.ProcessChoice();
}
CardGame savestate = gameIterator.game.Clone();
// THIS HELPS FIND ORIGINAL MOVE CHOICES
if (first)
{
if (!movestates.Keys.Contains(c))
{
movestates[c] = savestate;
}
first = false;
}
// Stateandplayer is Tuple with state after move, and the idx of the player who made the move
Tuple <CardGame, int> stateandplayer = Tuple.Create <CardGame, int>(savestate, idx);
// IF THIS IS THE FIRST SIMULATION WHICH HAS ARRIVED AT THIS STATE::
if (expand && (!plays.Keys.Contains(stateandplayer)))
{
expand = false;
plays[stateandplayer] = 0;
wins[stateandplayer] = 0;
movelist[c] = stateandplayer;
}
visitedstates.Add(stateandplayer);
}
else
{
gameIterator.ProcessChoice();
} // IF IT ISNT MY MOVE, DONT KEEP ANY DATA
}
// ProcessScore returns a sorted list
// where the winner is rank 0 for either min/max games.
var winners = gameIterator.ProcessScore();
double[] inverseRankSum = new double[numPlayers];
int p = 0;
foreach (Tuple <int, int> scoreandidx in winners)
{
inverseRankSum[scoreandidx.Item2] = ((double)1) / (p + 1);
p++;
}
// GO THROUGH VISITED STATES
foreach (Tuple <CardGame, int> stateandplayer in visitedstates)
{
if (plays.Keys.Contains(stateandplayer))
{
plays[stateandplayer] += 1;
wins[stateandplayer] += inverseRankSum[stateandplayer.Item2];
}
}
}