def info_best_move(self, action, value, depth):
self.end_time = time()
if not self.is_red_turn:
value = -value
score = int(value * 1000)
duration = self.end_time - self.start_time
nps = int(depth * 100 / duration) * 1000
print(f"info depth {depth} score {score} time {int(duration * 1000)} nps {nps}")
logger.debug(f"info depth {depth} score {score} time {int((self.end_time - self.start_time) * 1000)}")
sys.stdout.flush()
# get ponder
state = senv.step(self.state, action)
ponder = None
if state in self.search_tree:
node = self.search_tree[state]
cnt = 0
for mov, action_state in node.a.items():
if action_state.n > cnt:
ponder = mov
cnt = action_state.n
if not self.is_red_turn:
action = flip_move(action)
action = senv.to_uci_move(action)
output = f"bestmove {action}"
if ponder:
if self.is_red_turn:
ponder = flip_move(ponder)
ponder = senv.to_uci_move(ponder)
output += f" ponder {ponder}"
print(output)
logger.debug(output)
sys.stdout.flush()
def MCTS_search(self, env: CChessEnv, is_root_node=False, tid=0) -> float:
"""
Monte Carlo Tree Search
"""
if env.done:
if env.winner == Winner.draw:
return 0
else:
return -1
state = self.get_state_key(env)
with self.node_lock[state]:
if state not in self.tree:
# Expand and Evaluate
leaf_p, leaf_v = self.expand_and_evaluate(env)
# self.neural_net_out_p, self.neural_net_out_v = leaf_p, leaf_v
self.tree[state].p = leaf_p
self.tree[state].legal_moves = self.get_legal_moves(env)
return leaf_v
if tid in self.tree[state].visit: # loop
return 0
# Select
self.tree[state].visit.append(tid)
sel_action = self.select_action_q_and_u(state, is_root_node)
if sel_action is None:
return -1
virtual_loss = self.config.play.virtual_loss
self.tree[state].sum_n += virtual_loss
action_state = self.tree[state].a[sel_action]
action_state.n += virtual_loss
action_state.w -= virtual_loss
action_state.q = action_state.w / action_state.n
if env.red_to_move:
env.step(sel_action)
else:
env.step(flip_move(sel_action))
leaf_v = self.MCTS_search(env, not is_root_node, tid)
leaf_v = -leaf_v
# Backup
# update N, W, Q
with self.node_lock[state]:
node = self.tree[state]
node.visit.remove(tid)
node.sum_n = node.sum_n - virtual_loss + 1
action_state = node.a[sel_action]
action_state.n += 1 - virtual_loss
action_state.w += leaf_v + virtual_loss
action_state.q = action_state.w / action_state.n
return leaf_v
def get_catch_list(state, moves=None):
catch_list = set()
if not moves:
moves = get_legal_moves(state)
for mov in moves:
next_state, no_eat = new_step(state, mov)
if not no_eat: # 有吃子
# 判断能不能吃回来(防御)
could_defend = False
next_moves = get_legal_moves(next_state)
fliped_move = flip_move(mov)
dest = fliped_move[2:]
for nmov in next_moves:
if nmov[2:] == dest:
could_defend = True
break
if not could_defend:
i = int(mov[1])
j = int(mov[0])
black_board = state_to_board(state)
if black_board[i][j] == 'p' and i <= 4:
continue
m = int(mov[3])
n = int(mov[2])
if black_board[m][n] == 'P' and m > 4:
continue
# 判断是否为兑
if black_board[m][n].upper() == black_board[i][j].upper():
continue
# print(f"Catch: mov = {mov}, chessman = {black_board[i][j]}")
catch_list.add(
(black_board[i][j], i, j, black_board[m][n], m, n))
return catch_list
def test_static_env():
from cchess_alphazero.environment.env import CChessEnv
import cchess_alphazero.environment.static_env as senv
from cchess_alphazero.environment.static_env import INIT_STATE
from cchess_alphazero.environment.lookup_tables import flip_move
env = CChessEnv()
env.reset()
print("env: " + env.observation)
print("senv: " + INIT_STATE)
state = INIT_STATE
env.step('0001')
state = senv.step(state, '0001')
print(senv.evaluate(state))
print("env: " + env.observation)
print("senv: " + state)
env.step('7770')
state = senv.step(state, flip_move('7770'))
print(senv.evaluate(state))
print("env: " + env.observation)
print("senv: " + state)
env.render()
board = senv.state_to_board(state)
for i in range(9, -1, -1):
print(board[i])
print("env: ")
print(env.input_planes()[0+7:3+7])
print("senv: ")
print(senv.state_to_planes(state)[0+7:3+7])
print(f"env: {env.board.legal_moves()}" )
print(f"senv: {senv.get_legal_moves(state)}")
print(set(env.board.legal_moves()) == set(senv.get_legal_moves(state)))
def start(self, human_first=True):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config,
search_tree=defaultdict(VisitState),
pipes=self.pipe,
enable_resign=True,
debugging=False)
self.human_move_first = human_first
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
while not self.env.board.is_end():
if human_first == self.env.red_to_move:
self.env.board.calc_chessmans_moving_list()
is_correct_chessman = False
is_correct_position = False
chessman = None
while not is_correct_chessman:
title = "请输入棋子位置: "
input_chessman_pos = input(title)
x, y = int(input_chessman_pos[0]), int(
input_chessman_pos[1])
chessman = self.env.board.chessmans[x][y]
if chessman != None and chessman.is_red == self.env.board.is_red_turn:
is_correct_chessman = True
print(f"当前棋子为{chessman.name_cn}，可以落子的位置有：")
for point in chessman.moving_list:
print(point.x, point.y)
else:
print("没有找到此名字的棋子或未轮到此方走子")
while not is_correct_position:
title = "请输入落子的位置: "
input_chessman_pos = input(title)
x, y = int(input_chessman_pos[0]), int(
input_chessman_pos[1])
is_correct_position = chessman.move(x, y)
if is_correct_position:
self.env.board.print_to_cl()
self.env.board.clear_chessmans_moving_list()
else:
action, policy = self.ai.action(self.env.get_state(),
self.env.num_halfmoves)
if not self.env.red_to_move:
action = flip_move(action)
if action is None:
print("AI投降了!")
break
self.env.step(action)
print(f"AI选择移动 {action}")
self.env.board.print_to_cl()
self.ai.close()
print(f"胜者是 is {self.env.board.winner} !!!")
self.env.board.print_record()
def cmd_position(self):
'''
position {fen <fenstring> | startpos } [moves <move1> .... <moven>]
'''
if not self.is_ready:
return
move_idx = -1
if len(self.args) > 0:
if self.args[0] == 'fen':
# init with fen string
fen = self.args[1]
try:
self.state = senv.fen_to_state(fen)
except Exception as e:
logger.error(f"cmd position error! cmd = {self.args}, {e}")
return
self.history = [self.state]
turn = self.args[2]
if turn == 'b':
self.state = senv.fliped_state(self.state)
self.is_red_turn = False
self.turns = (int(self.args[6]) - 1) * 2 + 1
else:
self.is_red_turn = True
self.turns = (int(self.args[6]) - 1) * 2
if len(self.args) > 7 and self.args[7] == 'moves':
move_idx = 8
elif self.args[0] == 'startpos':
self.state = senv.INIT_STATE
self.is_red_turn = True
self.history = [self.state]
self.turns = 0
if len(self.args) > 1 and self.args[1] == 'moves':
move_idx = 2
elif self.args[0] == 'moves':
move_idx = 1
else:
self.state = senv.INIT_STATE
self.is_red_turn = True
self.history = [self.state]
self.turns = 0
logger.debug(f"state = {self.state}")
# senv.render(self.state)
# execute moves
if move_idx != -1:
for i in range(move_idx, len(self.args)):
action = senv.parse_ucci_move(self.args[i])
if not self.is_red_turn:
action = flip_move(action)
self.history.append(action)
self.state = senv.step(self.state, action)
self.is_red_turn = not self.is_red_turn
self.turns += 1
self.history.append(self.state)
logger.debug(f"state = {self.state}")
def start(self, human_first=True):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config, search_tree=defaultdict(VisitState), pipes=self.pipe,
enable_resign=True, debugging=False)
self.human_move_first = human_first
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
while not self.env.board.is_end():
if human_first == self.env.red_to_move:
self.env.board.calc_chessmans_moving_list()
is_correct_chessman = False
is_correct_position = False
chessman = None
while not is_correct_chessman:
title = "请输入棋子位置: "
input_chessman_pos = input(title)
x, y = int(input_chessman_pos[0]), int(input_chessman_pos[1])
chessman = self.env.board.chessmans[x][y]
if chessman != None and chessman.is_red == self.env.board.is_red_turn:
is_correct_chessman = True
print(f"当前棋子为{chessman.name_cn}，可以落子的位置有：")
for point in chessman.moving_list:
print(point.x, point.y)
else:
print("没有找到此名字的棋子或未轮到此方走子")
while not is_correct_position:
title = "请输入落子的位置: "
input_chessman_pos = input(title)
x, y = int(input_chessman_pos[0]), int(input_chessman_pos[1])
is_correct_position = chessman.move(x, y)
if is_correct_position:
self.env.board.print_to_cl()
self.env.board.clear_chessmans_moving_list()
else:
action, policy = self.ai.action(self.env.get_state(), self.env.num_halfmoves)
if not self.env.red_to_move:
action = flip_move(action)
if action is None:
print("AI投降了!")
break
self.env.step(action)
print(f"AI选择移动 {action}")
self.env.board.print_to_cl()
self.ai.close()
print(f"胜者是 is {self.env.board.winner} !!!")
self.env.board.print_record()
def test_sl():
from cchess_alphazero.worker import sl
from cchess_alphazero.config import Config
from cchess_alphazero.environment.lookup_tables import ActionLabelsRed, flip_policy, flip_move
c = Config('mini')
labels_n = len(ActionLabelsRed)
move_lookup = {move: i for move, i in zip(ActionLabelsRed, range(labels_n))}
slworker = sl.SupervisedWorker(c)
p1 = slworker.build_policy('0001', False)
print(p1[move_lookup['0001']])
p2 = slworker.build_policy('0001', True)
print(p2[move_lookup[flip_move('0001')]])
def test_onegreen():
import cchess_alphazero.environment.static_env as senv
from cchess_alphazero.environment.lookup_tables import flip_move
init = '9999299949999999249999869999999958999999519999999999999999997699'
state = senv.init(init)
print(state)
senv.render(state)
move = senv.parse_onegreen_move('8685')
state = senv.step(state, move)
print(state)
senv.render(state)
move = senv.parse_onegreen_move('7666')
state = senv.step(state, flip_move(move))
print(state)
senv.render(state)
def test_ucci():
import cchess_alphazero.environment.static_env as senv
from cchess_alphazero.environment.lookup_tables import flip_move
state = senv.INIT_STATE
state = senv.step(state, '0001')
fen = senv.state_to_fen(state, 1)
print(fen)
senv.render(state)
move = 'b7b0'
move = senv.parse_ucci_move(move)
print(f'Parsed move {move}')
move = flip_move(move)
print(f'fliped move {move}')
state = senv.step(state, move)
senv.render(state)
fen = senv.state_to_fen(state, 2)
print(fen)
def print_depth_info(self, state, turns, start_time, value, no_act):
'''
info depth xx pv xxx
'''
depth = self.done_tasks // 100
end_time = time()
pv = ""
i = 0
while i < 20:
node = self.tree[state]
bestmove = None
root = True
n = 0
if len(node.a) == 0:
break
for mov, action_state in node.a.items():
if action_state.n >= n:
if root and no_act and mov in no_act:
continue
n = action_state.n
bestmove = mov
if bestmove is None:
logger.error(
f"state = {state}, turns = {turns}, no_act = {no_act}, root = {root}, len(as) = {len(node.a)}"
)
break
state = senv.step(state, bestmove)
root = False
if turns % 2 == 1:
bestmove = flip_move(bestmove)
bestmove = senv.to_uci_move(bestmove)
pv += " " + bestmove
i += 1
turns += 1
if state in self.debug:
_, value = self.debug[state]
if turns % 2 != self.side:
value = -value
score = int(value * 1000)
duration = end_time - start_time
nps = int(depth * 100 / duration) * 1000
output = f"info depth {depth} score {score} time {int(duration * 1000)} pv" + pv + f" nps {nps}"
print(output)
logger.debug(output)
sys.stdout.flush()
def start(self):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config,
search_tree=defaultdict(VisitState),
pipes=self.pipe,
enable_resign=True,
debugging=False)
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
history = [self.env.get_state()]
while not self.env.board.is_end():
no_act = None
state = self.env.get_state()
if state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
no_act.append(history[i + 1])
action, _ = self.ai.action(state, self.env.num_halfmoves, no_act)
history.append(action)
if action is None:
print("AI投降了!")
break
move = self.env.board.make_single_record(int(action[0]),
int(action[1]),
int(action[2]),
int(action[3]))
if not self.env.red_to_move:
action = flip_move(action)
self.env.step(action)
history.append(self.env.get_state())
print(f"AI选择移动 {move}")
self.env.board.print_to_cl()
sleep(1)
self.ai.close()
print(f"胜者是 is {self.env.board.winner} !!!")
self.env.board.print_record()
def ai_move(self):
ai_move_first = not self.human_move_first
history = [self.env.get_state()]
no_act = None
while not self.env.done:
if ai_move_first == self.env.red_to_move:
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.ai.search_results = {}
state = self.env.get_state()
if state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
no_act.append(history[i + 1])
if no_act != []:
logger.debug(f"no_act = {no_act}")
action, policy = self.ai.action(state, self.env.num_halfmoves, no_act)
if action is None:
logger.info("AI has resigned!")
return
history.append(action)
if not self.env.red_to_move:
action = flip_move(action)
key = self.env.get_state()
p, v = self.ai.debug[key]
logger.info(f"NN value = {v:.3f}")
self.nn_value = v
logger.info("MCTS results:")
self.mcts_moves = {}
for move, action_state in self.ai.search_results.items():
move_cn = self.env.board.make_single_record(int(move[0]), int(move[1]), int(move[2]), int(move[3]))
logger.info(f"move: {move_cn}-{move}, visit count: {action_state[0]}, Q_value: {action_state[1]:.3f}, Prior: {action_state[2]:.3f}")
self.mcts_moves[move_cn] = action_state
x0, y0, x1, y1 = int(action[0]), int(action[1]), int(action[2]), int(action[3])
chessman_sprite = select_sprite_from_group(self.chessmans, x0, y0)
sprite_dest = select_sprite_from_group(self.chessmans, x1, y1)
if sprite_dest:
self.chessmans.remove(sprite_dest)
sprite_dest.kill()
chessman_sprite.move(x1, y1, self.chessman_w, self.chessman_h)
history.append(self.env.get_state())
def start(self):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config, search_tree=defaultdict(VisitState), pipes=self.pipe,
enable_resign=True, debugging=False)
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
history = [self.env.get_state()]
while not self.env.board.is_end():
no_act = None
state = self.env.get_state()
if state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
no_act.append(history[i + 1])
action, _ = self.ai.action(state, self.env.num_halfmoves, no_act)
history.append(action)
if action is None:
print("AI投降了!")
break
move = self.env.board.make_single_record(int(action[0]), int(action[1]), int(action[2]), int(action[3]))
if not self.env.red_to_move:
action = flip_move(action)
self.env.step(action)
history.append(self.env.get_state())
print(f"AI选择移动 {move}")
self.env.board.print_to_cl()
sleep(1)
self.ai.close()
print(f"胜者是 is {self.env.board.winner} !!!")
self.env.board.print_record()
def ai_move(self):
ai_move_first = not self.human_move_first
self.history = [self.env.get_state()]
no_act = None
while not self.env.done:
if ai_move_first == self.env.red_to_move:
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.ai.search_results = {}
state = self.env.get_state()
logger.info(f"state = {state}")
_, _, _, check = senv.done(state, need_check=True)
if not check and state in self.history[:-1]:
no_act = []
free_move = defaultdict(int)
for i in range(len(self.history) - 1):
if self.history[i] == state:
# 如果走了下一步是将军或捉：禁止走那步
if senv.will_check_or_catch(
state, self.history[i + 1]):
no_act.append(self.history[i + 1])
# 否则当作闲着处理
else:
free_move[state] += 1
if free_move[state] >= 2:
# 作和棋处理
self.env.winner = Winner.draw
self.env.board.winner = Winner.draw
break
if no_act:
logger.debug(f"no_act = {no_act}")
action, policy = self.ai.action(state, self.env.num_halfmoves,
no_act)
if action is None:
logger.info("AI has resigned!")
return
self.history.append(action)
if not self.env.red_to_move:
action = flip_move(action)
key = self.env.get_state()
p, v = self.ai.debug[key]
logger.info(f"check = {check}, NN value = {v:.3f}")
self.nn_value = v
logger.info("MCTS results:")
self.mcts_moves = {}
for move, action_state in self.ai.search_results.items():
move_cn = self.env.board.make_single_record(
int(move[0]), int(move[1]), int(move[2]), int(move[3]))
logger.info(
f"move: {move_cn}-{move}, visit count: {action_state[0]}, Q_value: {action_state[1]:.3f}, Prior: {action_state[2]:.3f}"
)
self.mcts_moves[move_cn] = action_state
x0, y0, x1, y1 = int(action[0]), int(action[1]), int(
action[2]), int(action[3])
chessman_sprite = select_sprite_from_group(
self.chessmans, x0, y0)
sprite_dest = select_sprite_from_group(self.chessmans, x1, y1)
if sprite_dest:
self.chessmans.remove(sprite_dest)
sprite_dest.kill()
chessman_sprite.move(x1, y1, self.chessman_w, self.chessman_h)
self.history.append(self.env.get_state())
def start(self):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config, search_tree=defaultdict(VisitState), pipes=self.pipe,
enable_resign=True, debugging=False)
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
history = [self.env.get_state()]
turns = 0
game_over = False
final_move = None
while not game_over:
if (self.ai_move_first and turns % 2 == 0) or (not self.ai_move_first and turns % 2 == 1):
start_time = time()
no_act = None
state = self.env.get_state()
if state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
act = history[i + 1]
if not self.env.red_to_move:
act = flip_move(act)
no_act.append(act)
action, _ = self.ai.action(state, self.env.num_halfmoves, no_act)
end_time = time()
if action is None:
print("AlphaZero 投降了!")
break
move = self.env.board.make_single_record(int(action[0]), int(action[1]), int(action[2]), int(action[3]))
print(f"AlphaZero 选择移动 {move}, 消耗时间 {(end_time - start_time):.2f}s")
if not self.env.red_to_move:
action = flip_move(action)
else:
state = self.env.get_state()
print(state)
fen = senv.state_to_fen(state, turns)
action = self.get_ucci_move(fen)
if action is None:
print("Eleeye 投降了!")
break
print(action)
if not self.env.red_to_move:
rec_action = flip_move(action)
else:
rec_action = action
move = self.env.board.make_single_record(int(rec_action[0]), int(rec_action[1]), int(rec_action[2]), int(rec_action[3]))
print(f"Eleeye 选择移动 {move}")
history.append(action)
self.env.step(action)
history.append(self.env.get_state())
self.env.board.print_to_cl()
turns += 1
sleep(1)
game_over, final_move = self.env.board.is_end_final_move()
print(game_over, final_move)
if final_move:
move = self.env.board.make_single_record(int(final_move[0]), int(final_move[1]), int(final_move[2]), int(final_move[3]))
print(f"Final Move {move}")
if not self.env.red_to_move:
final_move = flip_move(final_move)
self.env.step(final_move)
self.env.board.print_to_cl()
self.ai.close()
print(f"胜者是 is {self.env.board.winner} !!!")
self.env.board.print_record()
def start_game(self, idx, search_tree):
pipes = self.cur_pipes.pop()
if not self.config.play.share_mtcs_info_in_self_play or \
idx % self.config.play.reset_mtcs_info_per_game == 0:
search_tree = defaultdict(VisitState)
if random() > self.config.play.enable_resign_rate:
enable_resign = True
else:
enable_resign = False
self.player = CChessPlayer(self.config,
search_tree=search_tree,
pipes=pipes,
enable_resign=enable_resign,
debugging=False)
state = senv.INIT_STATE
history = [state]
value = 0
turns = 0 # even == red; odd == black
game_over = False
is_alpha_red = True if idx % 2 == 0 else False
final_move = None
check = False
while not game_over:
if (is_alpha_red and turns % 2 == 0) or (not is_alpha_red
and turns % 2 == 1):
no_act = None
if not check and state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
no_act.append(history[i + 1])
action, _ = self.player.action(state, turns, no_act)
if action is None:
logger.debug(
f"{turns % 2} (0 = red; 1 = black) has resigned!")
value = -1
break
else:
fen = senv.state_to_fen(state, turns)
action = self.get_ucci_move(fen)
if action is None:
logger.debug(
f"{turns % 2} (0 = red; 1 = black) has resigned!")
value = -1
break
if turns % 2 == 1:
action = flip_move(action)
history.append(action)
state = senv.step(state, action)
turns += 1
history.append(state)
if turns / 2 >= self.config.play.max_game_length:
game_over = True
value = 0
else:
game_over, value, final_move, check = senv.done(
state, need_check=True)
if final_move:
history.append(final_move)
state = senv.step(state, final_move)
history.append(state)
turns += 1
value = -value
self.player.close()
del search_tree
del self.player
gc.collect()
if turns % 2 == 1: # balck turn
value = -value
v = value
if turns <= 10:
if random() > 0.7:
store = True
else:
store = False
else:
store = True
if store:
data = [history[0]]
for i in range(turns):
k = i * 2
data.append([history[k + 1], value])
value = -value
self.save_play_data(idx, data)
self.cur_pipes.append(pipes)
self.remove_play_data()
return v, turns, state, store
def start_game(self, idx, search_tree):
pipes = self.cur_pipes.pop()
if not self.config.play.share_mtcs_info_in_self_play or \
idx % self.config.play.reset_mtcs_info_per_game == 0:
search_tree = defaultdict(VisitState)
if random() > self.config.play.enable_resign_rate:
enable_resign = True
else:
enable_resign = False
self.player = CChessPlayer(self.config, search_tree=search_tree, pipes=pipes, enable_resign=enable_resign, debugging=False)
state = senv.INIT_STATE
history = [state]
policys = []
value = 0
turns = 0 # even == red; odd == black
game_over = False
is_alpha_red = True if idx % 2 == 0 else False
final_move = None
while not game_over:
if (is_alpha_red and turns % 2 == 0) or (not is_alpha_red and turns % 2 == 1):
no_act = None
if state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
no_act.append(history[i + 1])
action, policy = self.player.action(state, turns, no_act)
if action is None:
logger.debug(f"{turns % 2} (0 = red; 1 = black) has resigned!")
value = -1
break
else:
fen = senv.state_to_fen(state, turns)
action = self.get_ucci_move(fen)
if action is None:
logger.debug(f"{turns % 2} (0 = red; 1 = black) has resigned!")
value = -1
break
if turns % 2 == 1:
action = flip_move(action)
try:
policy = self.build_policy(action, False)
except Exception as e:
logger.error(f"Build policy error {e}, action = {action}, state = {state}, fen = {fen}")
value = 0
break
history.append(action)
policys.append(policy)
state = senv.step(state, action)
turns += 1
history.append(state)
if turns / 2 >= self.config.play.max_game_length:
game_over = True
value = senv.evaluate(state)
else:
game_over, value, final_move = senv.done(state)
if final_move:
policy = self.build_policy(final_move, False)
history.append(final_move)
policys.append(policy)
state = senv.step(state, final_move)
history.append(state)
self.player.close()
if turns % 2 == 1: # balck turn
value = -value
v = value
if v == 0 or turns <= 10:
if random() > 0.7:
store = True
else:
store = False
else:
store = True
if store:
data = []
for i in range(turns):
k = i * 2
data.append([history[k], policys[i], value])
value = -value
self.save_play_data(idx, data)
self.cur_pipes.append(pipes)
self.remove_play_data()
return v, turns, state, search_tree, store
def start(self, human_first=True):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config,
search_tree=defaultdict(VisitState),
pipes=self.pipe,
enable_resign=True,
debugging=False)
self.human_move_first = human_first
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
while not self.env.board.is_end():
if human_first == self.env.red_to_move:
self.env.board.calc_chessmans_moving_list()
is_correct_chessman = False
is_correct_position = False
chessman = None
while not is_correct_chessman:
title = "Please enter the chess piece position: "
input_chessman_pos = input(title)
print(input_chessman_pos)
x, y = int(input_chessman_pos[0]), int(
input_chessman_pos[1])
chessman = self.env.board.chessmans[x][y]
if chessman != None and chessman.is_red == self.env.board.is_red_turn:
is_correct_chessman = True
print(
f"The current chess piece is {chessman.name}，places where you can play："
)
for point in chessman.moving_list:
print(point.x, point.y)
else:
print(
"No chess piece with this name was found or it was not his turn to walk"
)
while not is_correct_position:
title = "Please enter the location of the child: "
input_chessman_pos = input(title)
x, y = int(input_chessman_pos[0]), int(
input_chessman_pos[1])
is_correct_position = chessman.move(x, y)
if is_correct_position:
self.env.board.print_to_cl()
self.env.board.clear_chessmans_moving_list()
else:
action, policy = self.ai.action(self.env.get_state(),
self.env.num_halfmoves)
if not self.env.red_to_move:
action = flip_move(action)
if action is None:
print("AI surrendered!")
break
self.env.step(action)
print(f"AI chooses to move {action}")
self.env.board.print_to_cl()
self.ai.close()
print(f"The winner is is {self.env.board.winner} !!!")
self.env.board.print_record()
def start(self):
self.env.reset()
self.load_model()
self.pipe = self.model.get_pipes()
self.ai = CChessPlayer(self.config,
search_tree=defaultdict(VisitState),
pipes=self.pipe,
enable_resign=True,
debugging=False)
labels = ActionLabelsRed
labels_n = len(ActionLabelsRed)
self.env.board.print_to_cl()
history = [self.env.get_state()]
turns = 0
game_over = False
final_move = None
while not game_over:
if (self.ai_move_first
and turns % 2 == 0) or (not self.ai_move_first
and turns % 2 == 1):
start_time = time()
no_act = None
state = self.env.get_state()
if state in history[:-1]:
no_act = []
for i in range(len(history) - 1):
if history[i] == state:
act = history[i + 1]
if not self.env.red_to_move:
act = flip_move(act)
no_act.append(act)
action, _ = self.ai.action(state, self.env.num_halfmoves,
no_act)
end_time = time()
if action is None:
print("AlphaZero 投降了!")
break
move = self.env.board.make_single_record(
int(action[0]), int(action[1]), int(action[2]),
int(action[3]))
print(
f"AlphaZero 选择移动 {move}, 消耗时间 {(end_time - start_time):.2f}s"
)
if not self.env.red_to_move:
action = flip_move(action)
else:
state = self.env.get_state()
print(state)
fen = senv.state_to_fen(state, turns)
action = self.get_ucci_move(fen)
if action is None:
print("Eleeye 投降了!")
break
print(action)
if not self.env.red_to_move:
rec_action = flip_move(action)
else:
rec_action = action
move = self.env.board.make_single_record(
int(rec_action[0]), int(rec_action[1]), int(rec_action[2]),
int(rec_action[3]))
print(f"Eleeye 选择移动 {move}")
history.append(action)
self.env.step(action)
history.append(self.env.get_state())
self.env.board.print_to_cl()
turns += 1
sleep(1)
game_over, final_move = self.env.board.is_end_final_move()
print(game_over, final_move)
if final_move:
move = self.env.board.make_single_record(int(final_move[0]),
int(final_move[1]),
int(final_move[2]),
int(final_move[3]))
print(f"Final Move {move}")
if not self.env.red_to_move:
final_move = flip_move(final_move)
self.env.step(final_move)
self.env.board.print_to_cl()
self.ai.close()
print(f"胜者是 is {self.env.board.winner} !!!")
self.env.board.print_record()
