def genmove(self, moves, state, color):
assert not state.endOfGame()
moveNr = len(moves)
self.numSimulations = moveNr * 100
if moveNr == 1:
return moves[0]
#agent init
self.moves = moves
self.count = dict(zip(moves, [0] * moveNr))
self.avg_rewards = dict(zip(moves, [0] * moveNr))
#agent start
self.preAction = self._choose_action()
self.count[self.preAction] += 1
self.time += 1
coord = move_to_coord(self.preAction, state.size)
point = coord_to_point(coord[0], coord[1], state.size)
copy_board = copy.deepcopy(state)
copy_board.play_move_gomoku(point, color)
reward = copy_board.mysimulate(color)
self.avg_rewards[self.preAction] += (
(reward - self.avg_rewards[self.preAction]) /
self.count[self.preAction])
highest_reward = max(self.avg_rewards.values())
for move in self.avg_rewards:
if self.avg_rewards[move] == highest_reward:
self.bestMove = move
#agent step
for _ in range(self.numSimulations):
self.preAction = self._choose_action()
self.count[self.preAction] += 1
self.time += 1
coord = move_to_coord(self.preAction, state.size)
point = coord_to_point(coord[0], coord[1], state.size)
copy_board = copy.deepcopy(state)
copy_board.play_move_gomoku(point, color)
reward = copy_board.mysimulate(color)
self.avg_rewards[self.preAction] += (
(reward - self.avg_rewards[self.preAction]) /
self.count[self.preAction])
#update self.bestMove
if self.avg_rewards[self.preAction] > self.avg_rewards[
self.bestMove]:
self.bestMove = self.preAction
return self.bestMove
def play_move(board, move, color):
#print(type(move))
if isinstance(move,str):
coord = move_to_coord(move,board.size)
point = coord_to_point(coord[0],coord[1],board.size)
board.play_move_gomoku(point, color)
else:
board.play_move_gomoku(move, color)
def undo(board,move):
if isinstance(move,str):
coord = move_to_coord(move,board.size)
point = coord_to_point(coord[0],coord[1],board.size)
board.board[point]=EMPTY
board.current_player=GoBoardUtil.opponent(board.current_player)
else:
board.board[move]=EMPTY
board.current_player=GoBoardUtil.opponent(board.current_player)
def simulate(self, state, move, color):
stats = [0] * 3
#convert the last move to the index point
coord = move_to_coord(move, state.size)
point = coord_to_point(coord[0], coord[1], state.size)
state.play_move_gomoku(point, color)
moveNr = state.moveNumber()
for _ in range(self.numSimulations):
winner, _ = state.simulate()
stats[winner] += 1
state.resetToMoveNumber(moveNr)
assert sum(stats) == self.numSimulations
assert moveNr == state.moveNumber()
state.undoMove()
eval = (stats[BLACK] + 0.5 * stats[EMPTY]) / self.numSimulations
if state.current_player == WHITE:
eval = 1 - eval
return eval
