def self_play(env, agent, return_trajectory=False, verbose=False):
if return_trajectory:
trajectory = []
observation = env.reset()
for step in itertools.count():
board, player = observation
action, prob = agent.decide(observation, return_prob=True)
if verbose:
print(boardgame2.strfboard(board))
logging.info('第 {} 步：玩家 {}, 动作 {}'.format(step, player, action))
observation, winner, done, _ = env.step(action)
if return_trajectory:
trajectory.append((player, board, prob))
if done:
if verbose:
print(boardgame2.strfboard(observation[0]))
logging.info('赢家 {}'.format(winner))
break
if return_trajectory:
df_trajectory = pd.DataFrame(trajectory,
columns=['player', 'board', 'prob'])
df_trajectory['winner'] = winner
return df_trajectory
else:
return winner
def run(self, episodes=1, black_first=True, verbose=False):
black_win = 0
white_win = 0
for episode in range(1, episodes+1):
observation = self.env.reset()
done = False
turn = BLACK if black_first else WHITE
for step in range(1, 1000):
action = self.black_agent.decide(observation) if turn == BLACK else self.white_agent.decide(observation)
observation, winner, done, info = self.env.step(action)
if verbose:
print("第{}步, {}方落子: {}".format(step, "黑" if turn == BLACK else "白", action))
print(boardgame2.strfboard(observation[0]))
turn *= -1
if done:
break
if winner == BLACK:
black_win += 1
print("回合{}, 黑方胜利!".format(episode))
else:
white_win += 1
print("回合{}, 白方胜利!".format(episode))
self.env.close()
print("总计{}回合, 黑方胜率: {}, 白方胜率:{}".format(episode, black_win / episode, white_win / episode))
def search(self, board, prior_noise=False): # MCTS 搜索
s = boardgame2.strfboard(board)
if s not in self.winner:
self.winner[s] = self.env.get_winner((board, BLACK)) # 计算赢家
if self.winner[s] is not None: # 赢家确定的情况
return self.winner[s]
if s not in self.policy: # 未计算过策略的叶子节点
pis, vs = self.net.predict(board[np.newaxis])
pi, v = pis[0], vs[0]
valid = self.env.get_valid((board, BLACK))
masked_pi = pi * valid
total_masked_pi = np.sum(masked_pi)
if total_masked_pi <= 0: # 所有的有效动作都没有概率，偶尔可能发生
masked_pi = valid # workaround
total_masked_pi = np.sum(masked_pi)
self.policy[s] = masked_pi / total_masked_pi
self.valid[s] = valid
return v
# PUCT 上界计算
count_sum = self.count[s].sum()
coef = (self.c_init + np.log1p((1 + count_sum) / self.c_base)) * \
math.sqrt(count_sum) / (1. + self.count[s])
if prior_noise: # 先验噪声
alpha = 1. / self.valid[s].sum()
noise = np.random.gamma(alpha, 1., board.shape)
noise *= self.valid[s]
noise /= noise.sum()
prior = (1. - self.prior_exploration_fraction) * \
self.policy[s] + \
self.prior_exploration_fraction * noise
else:
prior = self.policy[s]
ub = np.where(self.valid[s], self.q[s] + coef * prior, np.nan)
location_index = np.nanargmax(ub)
location = np.unravel_index(location_index, board.shape)
(next_board, next_player), _, _, _ = self.env.next_step(
(board, BLACK), np.array(location))
next_canonical_board = next_player * next_board
next_v = self.search(next_canonical_board) # 递归搜索
v = next_player * next_v
self.count[s][location] += 1
self.q[s][location] += (v - self.q[s][location]) / \
self.count[s][location]
return v
def decide(self, observation, greedy=False, return_prob=False):
# 计算策略
board, player = observation
canonical_board = player * board
s = boardgame2.strfboard(canonical_board)
while self.count[s].sum() < self.sim_count: # 多次 MCTS 搜索
self.search(canonical_board, prior_noise=True)
prob = self.count[s] / self.count[s].sum()
# 采样
location_index = np.random.choice(prob.size, p=prob.reshape(-1))
location = np.unravel_index(location_index, prob.shape)
if return_prob:
return location, prob
return location
def hash_convert(self, state):
board, player = state
return (strfboard(board), player)
def __contains__(self, state):
board, player = state
board_ = board if isinstance(board, str) else strfboard(board)
return super().__contains__((board_, player))
def __setitem__(self, state, value):
board, player = state
board_ = board if isinstance(board, str) else strfboard(board)
return super().__setitem__((board_, player), value)