def __init__(self,
board_size=8,
random_rate=0.0,
as_player_2_rate=0.0,
use_agent_service=False):
self._agent = None # will need to set after Agent created
self._use_agent_service = use_agent_service
self._game: GameBoard = GameBoard(
board_size=board_size, random_start=random.random() < random_rate)
self._episode_ended = self._game.game_ended
self._player_id = PLAYER_1
self._log_on = False
self._random_rate = random_rate
self._as_player_2_rate = as_player_2_rate
self._action_spec = array_spec.BoundedArraySpec(
shape=(),
dtype=np.int32,
minimum=0,
# 0 - 8*8, including the last, for pass_turn
maximum=board_size * board_size,
name='action')
# self._observation_spec = array_spec.BoundedArraySpec(shape=(9,),
self._observation_spec = array_spec.BoundedArraySpec(
shape=(board_size, board_size, 1),
dtype=np.float32,
minimum=-1.0,
maximum=1.0,
name='observation')
def fight(policy_1, policy_2) -> GameBoard:
board_size = 8
board = GameBoard(board_size=board_size, random_start=False)
while not board.game_ended:
# player 1
board = step(board, PLAYER_1, policy_1)
if board.game_ended:
break
# player 2
board = step(board, PLAYER_2, policy_2)
return board
def _reset(self) -> ts:
self._game = GameBoard(
board_size=self._game.board_size,
random_start=random.random() < self._random_rate)
self._episode_ended = False
self._player_id = PLAYER_1
# random player_id
# if random.random() < self._random_rate:
if random.random() < self._as_player_2_rate:
self._player_id = PLAYER_2
# opponent move...
opponent_result = self._opponent_take_turn()
self._game = opponent_result.new_game_board
return ts.restart(self._get_observation(self._player_id))
def fight(policy, agent_as_player_id) -> GameBoard:
board_size = 8
board = GameBoard(board_size=board_size, random_start=False)
if agent_as_player_id == PLAYER_1:
while not board.game_ended:
# player 1
board = agent_step(board, PLAYER_1, policy)
if board.game_ended:
break
# player 2
board = server_step(board, PLAYER_2, PLAYER_1)
else:
while not board.game_ended:
# player 1
board = server_step(board, PLAYER_1, PLAYER_2)
if board.game_ended:
break
# player 2
board = agent_step(board, PLAYER_2, policy)
return board