def run(self):
self.ai = MCTSPlayer(c_puct=5, n_playout=20)
# self.ai.board = self.board
move = self.ai.get_action(self.board)
self.board.do_move(move)
print(move)
self.finishSignal.emit(int(move[0]), int(move[1]), int(move[2]),
int(move[3]))
class RL_QG_agent(object):
def __init__(self):
self.temp = 1e-3 # the temperature param
self.n_playout = 200 # num of simulations for each move
self.c_puct = 5
self.board_width = 8
self.board_height = 8
self.model_path = os.path.join("./models/curr_model_100rollout.pt")
#self.policy_value_net = PolicyValueNet(self.board_width, self.board_height, net_params=None)
#self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn, c_puct=self.c_puct, n_playout=self.n_playout)
self.mcts_player = MCTS_Pure(c_puct=5, n_playout=self.n_playout)
self.env = gym.make("Reversi8x8-v0")
self.init_model()
#self.load_model()
def init_model(self):
self.board = Board(env=self.env,
width=self.board_width,
height=self.board_height)
self.board.init_board()
self.game = Game(self.board)
self.have_step = False
def place(self, state, enables, player=None):
curr_state = bit_to_board(self.board.black, self.board.white)
curr_state = 1 - (curr_state[0] + curr_state[1])
reverse_change = np.where((curr_state - state[2]) == -1)
if self.have_step == False:
pass
elif reverse_change[0].shape[0] > 1:
self.board.init_board()
self.have_step = False
curr_state = bit_to_board(self.board.black, self.board.white)
curr_state = 1 - (curr_state[0] + curr_state[1])
change = np.where((curr_state - state[2]) == 1)
if change[0].shape[0] == 1:
action = change[0][0] * self.board_width + change[1][0]
self.board.do_move(action)
else:
if self.have_step == False:
pass
else:
action = 65
self.board.do_move(action)
move = self.mcts_player.get_action(self.board)
self.board.do_move(move)
self.have_step = True
return move
def load_model(self):
self.policy_value_net.policy_value_net.load_state_dict(
torch.load(self.model_path))
def run(states, sensible_moves, currentPlayer, lastMove):
n = 5
width, height = 8, 8
board = Board(width=width, height=height, n_in_row=n)
board.init_board()
board.states = states
board.availables = sensible_moves
board.current_player = currentPlayer
board.last_move = lastMove
#best_policy = PolicyValueNetNumpy(width, height, policy_param)
#mcts_player = MCTSPlayer(best_policy.policy_value_fn, c_puct=5, n_playout=400)
#只用纯MCTS
mcts_player = MCTS_Pure(c_puct=5, n_playout=4000) # n_playout参数 表示 搜索次数
nextmove = mcts_player.get_action(board)
return nextmove
def __init__(self):
self.temp = 1e-3 # the temperature param
self.n_playout = 200 # num of simulations for each move
self.c_puct = 5
self.board_width = 8
self.board_height = 8
self.model_path = os.path.join("./models/curr_model_100rollout.pt")
#self.policy_value_net = PolicyValueNet(self.board_width, self.board_height, net_params=None)
#self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn, c_puct=self.c_puct, n_playout=self.n_playout)
self.mcts_player = MCTS_Pure(c_puct=5, n_playout=self.n_playout)
self.env = gym.make("Reversi8x8-v0")
self.init_model()
class AI(QtCore.QThread):
finishSignal = QtCore.pyqtSignal(int, int, int, int)
# 构造函数里增加形参
def __init__(self, board, parent=None):
super(AI, self).__init__(parent)
self.board = board
# 重写 run() 函数
def run(self):
self.ai = MCTSPlayer(c_puct=5, n_playout=20)
# self.ai.board = self.board
move = self.ai.get_action(self.board)
self.board.do_move(move)
print(move)
self.finishSignal.emit(int(move[0]), int(move[1]), int(move[2]),
int(move[3]))
opponent.reset_player()
board = BoardSlim(width=BOARD_SHAPE[0],
height=BOARD_SHAPE[1],
n_in_row=4)
game = Game(board)
winner = game.start_play(player,
opponent,
start_player=i % 2,
is_shown=(i == 0) and show_game)
print("winner: ", winner)
if winner == 1:
stats["wins"] += 1
elif winner == -1:
stats["ties"] += 1
stats["length"].append(BOARD_SHAPE[0] * BOARD_SHAPE[1] -
len(board.availables))
return stats
if __name__ == "__main__":
player = load_player()
for playout in playouts:
mcts_player = MCTSPlayer(n_playout=playout)
stats = evaluate_player(player, mcts_player, show_game=True)
print()
print("win ratio agains %d playouts: " % playout,
stats["wins"] / num_games)
print("tie ratio agains %d playouts: " % playout,
stats["ties"] / num_games)
print("average game length agains %d playouts: ",
np.mean(stats["length"]))
opts, args = getopt.getopt(sys.argv[1:], "c:m:")
except getopt.GetoptError:
print('TwelveShogi.py -c <startColor(0 or 1)> -m <mode(0 or 1 or 2)>')
sys.exit()
for opt, arg in opts:
if opt == '-c':
startColor = int(arg)
elif opt == '-m':
mode = int(arg)
shogiboard = ShogiBoard()
shogiboard.curStepColor = startColor
shogiboard.redrawBoard(window)
if mode == 1:
# AIPlayer = TSAI()
AIPlayer = MCTS_Pure()
# AIPlayer = MCTSPlayer()
elif mode == 2:
# AIPlayer = TSAI()
AIPlayer = MCTS_Pure()
trainCounts = 0
# AIPlayer = MCTSPlayer(is_selfplay = 1)
top = 80
left = 70
yGap = 125
xGap = 140
curRow = 3
curCol = 0