def run_single_game(self):
board = CheckerBoard()
turn = 0
unresolved = False
while not board.is_over():
turn += 1
log_file.write("#### Turn %3d\n" % turn)
log_file.write(str(board))
log_file.flush()
if turn % 100 == 0:
print("Over %d turns played" % turn)
for player, agent in self.players.items():
while not board.is_over() and board.active == player:
print("Player %d is making a decision" % player)
start_time = time.time()
move = agent.best_move(board)
self.stats["thinking_time"][player].append(time.time() -
start_time)
board.update(move)
if turn > 200:
unresolved = True
break
self.stats["score"].append(board.winner if not unresolved else -1)
self.stats["played_rounds"] += turn
class CheckersEnvironmentWrapper:
def __init__(self):
# env initialization
self.actions = {}
self.observation = []
self.reward = 0
self.done = False
self.last_action_idx = 0
# initialize the board
self.board = CheckerBoard()
self.width = len(self.board.get_state_vector())
self.height = 1
self.win_reward = 100
self.defeat_reward = -100
self.game_turns = 0
self.score = 0
self.enable_capturing_reward = False
for idx, move in enumerate(self.board.get_all_moves()):
self.actions[idx] = move
print("total actions: ", len(self.actions))
self.action_space_size = len(self.actions)
self.reset()
def update_game_info(self):
self.observation = self.board.get_state_vector()
def restart_environment_episode(self):
self.board = CheckerBoard()
self.update_game_info()
return self.observation
def _idx_to_action(self, action_idx):
return self.actions[action_idx]
def get_valid_idx_actions(self):
possible_idx_actions = []
possible_moves = self.board.get_legal_moves()
for idx, action in self.actions.items():
if action in possible_moves:
possible_idx_actions.append(idx)
return possible_idx_actions
def step(self, action_idx):
assert self.board.get_current_player(
) == self.board.BLACK_PLAYER, "Training player should be black!"
self.last_action_idx = action_idx
action = self.actions[action_idx]
# print("take action ", action_idx, " : ", action)
white_pieces_before = self.board.get_white_num(
) + self.board.get_white_kings_num()
white_kings_pieces_before = self.board.get_white_kings_num()
black_pieces_before = self.board.get_black_num(
) + self.board.get_black_kings_num()
black_kings_pieces_before = self.board.get_black_kings_num()
self.board.make_move(action)
if self.board.get_current_player() == self.board.WHITE_PLAYER:
if not self.board.is_over():
# make AI opponent move
self.opponent_move()
self.update_game_info()
white_pieces = self.board.get_white_num(
) + self.board.get_white_kings_num()
white_kings_pieces = self.board.get_white_kings_num()
black_pieces = self.board.get_black_num(
) + self.board.get_black_kings_num()
black_kings_pieces = self.board.get_black_kings_num()
if self.board.is_over():
print("black: p. %d, k. %d, white: p. %d, k. %d" %
(black_pieces, black_kings_pieces, white_pieces,
white_kings_pieces))
if self.board.get_winner() == self.board.BLACK_PLAYER:
# black wins
print("black wins")
self.reward = self.win_reward
else:
print("white wins")
self.reward = self.defeat_reward
else:
if self.enable_capturing_reward:
captured_whites = white_pieces_before - white_pieces
captured_black = black_pieces_before - black_pieces
self.reward = captured_whites - captured_black
else:
self.reward = 0
self.score += self.reward
self.game_turns += 1
self.done = self.board.is_over()
return self.observation, self.reward, self.done
def opponent_move(self):
current_player = self.board.get_current_player()
moves = self.board.get_legal_moves()
action = random.choice(moves)
# print("opponent takes action ", action)
self.board.make_move(action)
if self.board.get_current_player() == current_player:
# print("opponent takes a jump")
self.opponent_move()
def reset(self):
self.restart_environment_episode()
self.done = False
self.reward = 0.0
self.last_action_idx = 0
self.game_turns = 0
self.score = 0
return self.observation, self.reward, self.done