// Starts the time limited Monte Carlo Tree Search and returns the best child node
// resulting from the search
public Node TimeLimitedMCTS(State rootState, int timeLimit)
{
Stopwatch timer = new Stopwatch();
Node bestNode = null;
while (bestNode == null && !rootState.IsGameOver())
{
timer.Start();
Node rootNode = TimeLimited(rootState, timeLimit, timer);
bestNode = FindBestChild(rootNode.Children);
timeLimit += 10;
timer.Reset();
}
return bestNode;
}
// recursive part of the minimax algorithm when used in iterative deepening search
// checks at each recursion if timeLimit has been reached
// if is has, it cuts of the search and returns the best move found so far, along with a boolean indicating that the search was not fully completed
private Tuple<Move, Boolean> IterativeDeepeningAlphaBeta(State state, int depth, double alpha, double beta, double timeLimit, Stopwatch timer)
{
Move bestMove;
if (depth == 0 || state.IsGameOver())
{
if (state.Player == GameEngine.PLAYER)
{
bestMove = new PlayerMove(); // dummy action, as there will be no valid move
bestMove.Score = AI.Evaluate(state);
return new Tuple<Move, Boolean>(bestMove, true);
}
else if (state.Player == GameEngine.COMPUTER)
{
bestMove = new ComputerMove(); // dummy action, as there will be no valid move
bestMove.Score = AI.Evaluate(state);
return new Tuple<Move, Boolean>(bestMove, true);
}
else throw new Exception();
}
if (state.Player == GameEngine.PLAYER) // AI's turn
{
bestMove = new PlayerMove();
double highestScore = Double.MinValue, currentScore = Double.MinValue;
List<Move> moves = state.GetMoves();
foreach (Move move in moves)
{
State resultingState = state.ApplyMove(move);
currentScore = IterativeDeepeningAlphaBeta(resultingState, depth - 1, alpha, beta, timeLimit, timer).Item1.Score;
if (currentScore > highestScore)
{
highestScore = currentScore;
bestMove = move;
}
alpha = Math.Max(alpha, highestScore);
if (beta <= alpha)
{ // beta cut-off
break;
}
if (timer.ElapsedMilliseconds > timeLimit)
{
bestMove.Score = highestScore;
return new Tuple<Move, Boolean>(bestMove, false); // recursion not completed, return false
}
}
bestMove.Score = highestScore;
return new Tuple<Move, Boolean>(bestMove, true);
}
else if (state.Player == GameEngine.COMPUTER) // computer's turn (the random event node)
{
bestMove = new ComputerMove();
double lowestScore = Double.MaxValue, currentScore = Double.MaxValue;
List<Move> moves = state.GetMoves();
foreach (Move move in moves)
{
State resultingState = state.ApplyMove(move);
currentScore = IterativeDeepeningAlphaBeta(resultingState, depth - 1, alpha, beta, timeLimit, timer).Item1.Score;
if (currentScore < lowestScore)
{
lowestScore = currentScore;
bestMove = move;
}
beta = Math.Min(beta, lowestScore);
if (beta <= alpha)
break;
if (timer.ElapsedMilliseconds > timeLimit)
{
bestMove.Score = lowestScore;
return new Tuple<Move, Boolean>(bestMove, false); // recursion not completed, return false
}
}
bestMove.Score = lowestScore;
return new Tuple<Move, Boolean>(bestMove, true);
}
else throw new Exception();
}
// Recursive part of iterative deepening Expectimax
private Tuple<Move, Boolean> IterativeDeepeningExpectimax(State state, int depth, double timeLimit, Stopwatch timer)
{
Move bestMove;
if (depth == 0 || state.IsGameOver())
{
if (state.Player == GameEngine.PLAYER)
{
bestMove = new PlayerMove(); // dummy action, as there will be no valid move
bestMove.Score = AI.Evaluate(state);
return new Tuple<Move, Boolean>(bestMove, true);
}
else if (state.Player == GameEngine.COMPUTER)
{
bestMove = new ComputerMove(); // dummy action, as there will be no valid move
bestMove.Score = AI.Evaluate(state);
return new Tuple<Move, Boolean>(bestMove, true);
}
else throw new Exception();
}
if (state.Player == GameEngine.PLAYER) // AI's turn
{
bestMove = new PlayerMove();
double highestScore = Double.MinValue, currentScore = Double.MinValue;
List<Move> moves = state.GetMoves();
foreach (Move move in moves)
{
State resultingState = state.ApplyMove(move);
currentScore = IterativeDeepeningExpectimax(resultingState, depth - 1, timeLimit, timer).Item1.Score;
if (currentScore > highestScore)
{
highestScore = currentScore;
bestMove = move;
}
if (timer.ElapsedMilliseconds > timeLimit)
{
bestMove.Score = highestScore;
return new Tuple<Move, Boolean>(bestMove, false); // recursion not completed, return false
}
}
bestMove.Score = highestScore;
return new Tuple<Move, Boolean>(bestMove, true);
}
else if (state.Player == GameEngine.COMPUTER) // computer's turn (the random event node)
{
bestMove = new ComputerMove();
// return the weighted average of all the child nodes's scores
double average = 0;
List<Cell> availableCells = state.GetAvailableCells();
List<Move> moves = state.GetAllComputerMoves(availableCells);
int moveCheckedSoFar = 0;
foreach (Move move in moves)
{
State resultingState = state.ApplyMove(move);
average += StateProbability(((ComputerMove)move).Tile) * IterativeDeepeningExpectimax(resultingState, depth - 1, timeLimit, timer).Item1.Score;
moveCheckedSoFar++;
if (timer.ElapsedMilliseconds > timeLimit)
{
bestMove.Score = average / moveCheckedSoFar;
return new Tuple<Move, Boolean>(bestMove, false); // recursion not completed, return false
}
}
bestMove.Score = average / moves.Count;
return new Tuple<Move, Boolean>(bestMove, true);
}
else throw new Exception();
}