private double HeuristicEval(BoardStates player, OthelloGame game)
{
//Based of features of the board that humans have identified.
//Hints of evaluation from any source I could find
//idealy these could me optimized using a genetic algorithm,
//but that is a different project
const int searchableDepthOverride = 2; //override min-max in favor of complete evaluation
const int endGame = 20; //<20 moves is endgame
const int midGame = 40; // 20 moves in is midgame
double value = 0;
int empty = game.GetPieceCount(BoardStates.empty);
if (game.GameComplete)
{
return(CompleteEval(player, game));
}
else if (empty < searchableDepthOverride)
{
return(MinimaxAlphaBeta(game, searchableDepthOverride, int.MinValue, int.MaxValue, player));
}
value += coinDiffWeight * Math.Pow((game.GetPieceCount(player) - game.GetPieceCount(~player) + empty - coinDiffOffset), coinDiffPower);
value += cornerDiffWeight * Math.Pow((game.GetCornerCount(player) - game.GetCornerCount(~player) + empty - cornerDiffOffset), cornerDiffPower);
value += nearCornerDiffWeight * Math.Pow((game.GetAdjCornerCount(player) - game.GetAdjCornerCount(~player) + empty - nearCornerDiffOffset), nearCornerDiffPower);
value += avalibleMoveDiffWeight * Math.Pow((game.GetPossiblePlayList(player).Count() - game.GetPossiblePlayList(~player).Count() + empty - avalibleMoveDiffOffset), avalibleMoveDiffPower);
value += nonTurnableCoinDiffWeight * Math.Pow((game.GetSafePeiceCountEstimation(player) - game.GetSafePeiceCountEstimation(~player) + empty - nonTurnableCoinDiffOffset), nonTurnableCoinDiffPower);
value += ControlledCornerDiffWeight * Math.Pow((game.GetControlledCorners(player) - game.GetControlledCorners(~player) + empty - ControlledCornerDiffOffset), ControlledCornerDiffPower);
return(value);
}
private byte[] PredictBestMove(int depth, OthelloGame game, BoardStates player)
{
byte[] bestMove = new byte[] { byte.MaxValue, byte.MaxValue };
List <byte[]> moves = game.GetPossiblePlayList();
double bestScore = int.MinValue + 1;
if (game.GetPieceCount(BoardStates.empty) > 58)//first two moves, don't compute
{
return(OpeningMove(player, game));
}
else if (moves.Count == 1) //don't compute if there is only 1 move
{
return(moves[0]);
}
foreach (byte[] move in moves)
{
OthelloGame testGame = game.DeepCopy();
testGame.MakeMove(move);
double thisScore = MinimaxAlphaBeta(testGame, depth - 1, double.MinValue, double.MaxValue, player);
if (thisScore > bestScore)
{
bestScore = thisScore;
bestMove = move;
}
}
return(bestMove);
}
private static byte[] OpeningMove(BoardStates player, OthelloGame game)
{//avoid computation for first move - only one symmetric option
//randomly select perpendicular or diagonal for second move - parallel
//has been shown to be much worse
//SPECIFIC TO 8x8 BOARDS
byte[][] firstMoves = new byte[4][] {
new byte[] { 2, 3 },
new byte[] { 3, 2 },
new byte[] { 4, 5 },
new byte[] { 5, 4 }
};
if (game.GetPieceCount(BoardStates.empty) == 60)
{
Random rndGen = new Random();
int rand = (int)Math.Ceiling(rndGen.NextDouble() * 4);
switch (rand)
{
case 1:
return(firstMoves[0]);
case 2:
return(firstMoves[1]);
case 3:
return(firstMoves[2]);
case 4:
return(firstMoves[3]);
default:
throw new Exception("OpeningMove has faulted with random number generation");
}
}
if (game.GetPieceCount(BoardStates.empty) == 59)
{
List <byte[]> moves = game.GetPossiblePlayList();
Random rndGen = new Random();
byte rand = (byte)Math.Ceiling(rndGen.NextDouble() * 2);
switch (rand)
{
case 1: //diagonal
return(moves[0]);
case 2: //perpendicular
return(moves[0]);
default:
throw new Exception("Opening move has faulted with random number generation");
}
}
return(new byte[] { byte.MaxValue, byte.MaxValue });
}
private double MinimaxAlphaBeta(OthelloGame board, int depth, double a, double b, BoardStates player)// bool isMaxPlayer)
{
// The heart of our AI. Minimax algorithm with alpha-beta pruning to speed up computation.
// Higher search depths = greater difficulty.
//from oliverzh200/reversi https://github.com/oliverzh2000/reversi
if (depth == 0 || board.GameComplete)
{
return(HeuristicEval(player, board));
}
double bestScore = double.MinValue;
List <byte[]> validMoves = board.GetPossiblePlayList();
if (validMoves.Count > 0)
{
foreach (byte[] move in validMoves)
{
OthelloGame childBoard = board.DeepCopy();
childBoard.MakeMove(move);
double nodeScore = MinimaxAlphaBeta(childBoard, depth - 1, a, b, player);
bestScore = Math.Max(bestScore, nodeScore);
a = Math.Max(bestScore, a);
if (b <= a) //Prune
{
break;
}
}
}
else
{
return(MinimaxAlphaBeta(board, depth, a, b, player));
}
return(bestScore);
}
private byte[] PredictBestMove(int depth, OthelloGame game, BoardStates player)
{
byte[] bestMove = new byte[] { byte.MaxValue, byte.MaxValue };
List <byte[]> moves = game.GetPossiblePlayList();
double bestScore = int.MinValue + 1;
foreach (byte[] move in moves)
{
OthelloGame testGame = game.DeepCopy();
testGame.MakeMove(move);
double thisScore = MinimaxAlphaBeta(testGame, depth - 1, double.MinValue, double.MaxValue, player);
if (thisScore > bestScore)
{
bestScore = thisScore;
bestMove = move;
}
}
if ((bestMove[0] == byte.MaxValue || bestMove[1] == byte.MaxValue) && moves.Count > 0)
{//All moves are valued at -inf, return one of em
return(moves[0]);
}
return(bestMove);
}