public Action GetActionFromRule(MinesweeperRule mr)
{
height = mr.height;
width = mr.width;
booms = mr.booms;
int[][] curState = mr.GetCurrentState();
int[][] aiView = new int[height][];
List <Position> blankList = new List <Position>();
List <Position> numberList = new List <Position>();
List <Position> blankBoundary = new List <Position>();
List <Position> unknownBlank = new List <Position>();
Dictionary <Position, List <Position> > numberNearBlank = new Dictionary <Position, List <Position> >();
Dictionary <Position, List <Position> > blankNearNumber = new Dictionary <Position, List <Position> >();
int remainBoom = mr.booms - mr.totalFlag;
Random rd = new Random();
for (int i = 0; i < height; i++)
{
aiView[i] = new int[width];
for (int j = 0; j < width; j++)
{
if (curState[i][j] == MinesweeperRule.MS_blank || curState[i][j] == MinesweeperRule.MS_bombquestion)
{
aiView[i][j] = AI_blank;
blankList.Add(new Position(i, j));
}
else if (curState[i][j] >= MinesweeperRule.MS_open0 && curState[i][j] <= MinesweeperRule.MS_open8)
{
numberList.Add(new Position(i, j));
int boomAround = curState[i][j] - MinesweeperRule.MS_open0 - mr.CountBoomAround(i, j, curState, MinesweeperRule.MS_bombflagged);
if (boomAround > 0)
{
aiView[i][j] = boomAround;
}
else
{
aiView[i][j] = AI_open0;
}
}
else
{
aiView[i][j] = AI_wall;
}
}
}
if (blankList.Count == height * width)
{
return(new Action {
x = rd.Next(height), y = rd.Next(width), mouse = 0
});
}
if (blankList.Count == 0)
{
return(new Action {
x = 0, y = 0, mouse = -1
});
}
for (int i = 0; i < height; i++)
{
for (int j = 0; j < width; j++)
{
if (aiView[i][j] == AI_blank)
{
Position curPos = new Position(i, j);
List <Position> nearNumbers = new List <Position>();
Position pos;
if (numberList.Contains(pos = new Position(i - 1, j - 1)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i - 1, j)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i - 1, j + 1)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i, j - 1)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i, j + 1)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i + 1, j - 1)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i + 1, j)))
{
nearNumbers.Add(pos);
}
if (numberList.Contains(pos = new Position(i + 1, j + 1)))
{
nearNumbers.Add(pos);
}
if (nearNumbers.Count > 0)
{
blankBoundary.Add(curPos);
numberNearBlank[curPos] = nearNumbers;
foreach (Position num in nearNumbers)
{
if (!blankNearNumber.ContainsKey(num))
{
List <Position> nearBlanks = new List <Position>();
if (blankList.Contains(pos = new Position(num.x - 1, num.y - 1)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x - 1, num.y)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x - 1, num.y + 1)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x, num.y - 1)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x, num.y + 1)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x + 1, num.y - 1)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x + 1, num.y)))
{
nearBlanks.Add(pos);
}
if (blankList.Contains(pos = new Position(num.x + 1, num.y + 1)))
{
nearBlanks.Add(pos);
}
blankNearNumber[num] = nearBlanks;
}
}
}
else
{
unknownBlank.Add(new Position(i, j));
}
}
}
}
foreach (var pos in blankBoundary)
{
foreach (var num in numberNearBlank[pos])
{
if (blankNearNumber[num].Count == aiView[num.x][num.y])
{
return(new Action {
x = pos.x, y = pos.y, mouse = 1
}); // right click to pos
}
else if (AI_open0 == aiView[num.x][num.y])
{
return(new Action {
x = pos.x, y = pos.y, mouse = 0
}); // left click to pos
}
}
}
Dictionary <Position, double> estimations = blankBoundary.ToDictionary(x => x, x => 0.0);
List <List <Position> > groupPosList = new List <List <Position> >();
List <Position> posList = new List <Position>();
Queue <Position> groupPosQue = new Queue <Position>();
if (blankBoundary.Count > 0)
{
groupPosQue.Enqueue(blankBoundary.First());
}
while (blankBoundary.Count > 0 && groupPosQue.Count > 0)
{
var pos = groupPosQue.Dequeue();
posList.Add(pos);
blankBoundary.Remove(pos);
foreach (var num in numberNearBlank[pos])
{
foreach (var blank in blankNearNumber[num])
{
if (blankBoundary.Contains(blank) && !groupPosQue.Contains(blank))
{
groupPosQue.Enqueue(blank);
}
}
}
if (groupPosQue.Count == 0)
{
groupPosList.Add(posList);
posList = new List <Position>();
if (blankBoundary.Count > 0)
{
groupPosQue.Enqueue(blankBoundary.First());
}
}
}
double placedFlag = 0;
double alpha = 0.1;
double epsilon = 0.05;
double addition = 0;
double max = -1000;
double est = 0;
double gama = 0.9;
foreach (var groupPos in groupPosList)
{
List <Position> numList = new List <Position>();
foreach (var pos in groupPos)
{
foreach (var num in numberNearBlank[pos])
{
if (!numList.Contains(num))
{
numList.Add(num);
}
}
}
List <double> placedFlagList = new List <double>();
int enchos = groupPos.Count * 20;
for (int count = 0; count < enchos; count++)
{
if (count < enchos * 3 / 4)
{
epsilon = 0.05;
}
else
{
epsilon = 0;
}
List <Position> tempGroupPos = groupPos.ToList();
Dictionary <Position, int> actions = groupPos.ToDictionary(x => x, x => 0); // 1: place flag, -1: don't place flag, 0: not yet action
List <Position> positiveAction = new List <Position>();
Dictionary <Position, int> numOfBlankAround = blankNearNumber.ToDictionary(x => x.Key, x => x.Value.Count);
int[][] tempAiView = aiView.Select(x => x.ToArray()).ToArray();
bool isFirst = true;
while (true)
{
bool checkWrong = false;
for (int i = 0; i < numList.Count; i++)
{
if (tempAiView[numList[i].x][numList[i].y] < 0 || tempAiView[numList[i].x][numList[i].y] > numOfBlankAround[numList[i]])
{
checkWrong = true;
break;
}
}
if (checkWrong)
{
gama = 1;
// update estimatios when wrong
for (int i = 0; i < positiveAction.Count; i++)
{
var pos = positiveAction[i];
estimations[pos] += alpha * gama * ((-1) * actions[pos] - estimations[pos]);
gama *= 0.8;
}
break;
}
if (!actions.Values.Contains(0))
{
// update estimatios when success
int hs = 1;
if (estimations.Values.Where(x => x == 1).Count() > remainBoom)
{
hs = -1;
}
gama = 1;
for (int i = 0; i < positiveAction.Count; i++)
{
var pos = positiveAction[i];
estimations[pos] += alpha * gama * (hs * actions[pos] - estimations[pos]);
gama *= 0.8;
}
placedFlagList.Add(actions.Values.Where(x => x == 1).Count());
break;
}
List <Position> needNoFlagList = new List <Position>();
foreach (var num in numList)
{
if (tempAiView[num.x][num.y] == 0)
{
foreach (var blank in blankNearNumber[num])
{
if (actions[blank] == 0 && !needNoFlagList.Contains(blank))
{
needNoFlagList.Add(blank);
}
}
}
}
if (needNoFlagList.Count > 0)
{
foreach (var blank in needNoFlagList)
{
actions[blank] = -1;
tempGroupPos.Remove(blank);
foreach (var num in numberNearBlank[blank])
{
numOfBlankAround[num]--;
}
}
continue;
}
List <Position> needFlagList = new List <Position>();
foreach (var num in numList)
{
if (tempAiView[num.x][num.y] == numOfBlankAround[num])
{
foreach (var blank in blankNearNumber[num])
{
if (actions[blank] == 0 && !needFlagList.Contains(blank))
{
needFlagList.Add(blank);
}
}
}
}
if (needFlagList.Count > 0)
{
foreach (var blank in needFlagList)
{
actions[blank] = 1;
tempGroupPos.Remove(blank);
foreach (var num in numberNearBlank[blank])
{
numOfBlankAround[num]--;
tempAiView[num.x][num.y]--;
}
}
continue;
}
if (isFirst)
{
Position pos = groupPos[rd.Next(groupPos.Count)];
int act = rd.Next(2) * 2 - 1;
actions[pos] = act;
tempGroupPos.Remove(pos);
positiveAction.Add(pos);
foreach (var num in numberNearBlank[pos])
{
numOfBlankAround[num]--;
if (act == 1)
{
tempAiView[num.x][num.y]--;
}
}
isFirst = false;
}
else
{
if (rd.NextDouble() < epsilon)
{
Position pos = tempGroupPos[rd.Next(tempGroupPos.Count)];
int act = rd.Next(2) * 2 - 1;
actions[pos] = act;
tempGroupPos.Remove(pos);
positiveAction.Add(pos);
foreach (var num in numberNearBlank[pos])
{
numOfBlankAround[num]--;
if (act == 1)
{
tempAiView[num.x][num.y]--;
}
}
}
else
{
List <Position> bestChoiceList = new List <Position>();
max = -1000;
foreach (var pos in tempGroupPos)
{
est = estimations[pos] < 0 ? -estimations[pos] : estimations[pos];
if (est > max)
{
max = est;
bestChoiceList = new List <Position>();
bestChoiceList.Add(pos);
}
else if (est == max)
{
bestChoiceList.Add(pos);
}
}
Position bestChoice = bestChoiceList[rd.Next(bestChoiceList.Count)];
int act = estimations[bestChoice] >= 0 ? 1 : -1;
actions[bestChoice] = act;
tempGroupPos.Remove(bestChoice);
positiveAction.Add(bestChoice);
foreach (var num in numberNearBlank[bestChoice])
{
numOfBlankAround[num]--;
if (act == 1)
{
tempAiView[num.x][num.y]--;
}
}
}
}
}
}
if (placedFlagList.Count > 0)
{
placedFlag += placedFlagList.Sum() * 1.0 / placedFlagList.Count;
}
else
{
addition += groupPos.Count;
}
}
List <Position> bestPosList = new List <Position>();
max = -1000;
foreach (var pos in estimations.Keys)
{
est = estimations[pos] < 0 ? -estimations[pos] : estimations[pos];
if (est > max)
{
max = est;
bestPosList = new List <Position>();
bestPosList.Add(pos);
}
else if (est == max)
{
bestPosList.Add(pos);
}
}
Position bestPos = new Position(0, 0);
if (bestPosList.Count > 0)
{
bestPos = bestPosList[rd.Next(bestPosList.Count)];
}
est = 0;
if (unknownBlank.Count > 0)
{
est = (remainBoom - placedFlag) * 1.0 / (unknownBlank.Count + addition) * 2 - 1;
foreach (var blk in unknownBlank)
{
estimations[blk] = est;
}
}
if (max < 0.9 && est > max)
{
bestPos = unknownBlank[rd.Next(unknownBlank.Count)];
}
return(new Action {
x = bestPos.x, y = bestPos.y, mouse = estimations[bestPos] >= 0 ? 1 : 0
});
}