def __init__(self, config):
r"""Creates an environment with the given game configuration.
Args:
config: dict, With parameters for the game. Config takes the following
keys and values.
- colors: int, Number of colors \in [2,5].
- ranks: int, Number of ranks \in [2,5].
- players: int, Number of players \in [2,5].
- hand_size: int, Hand size \in [4,5].
- max_information_tokens: int, Number of information tokens (>=0).
- max_life_tokens: int, Number of life tokens (>=1).
- observation_type: int.
0: Minimal observation.
1: First-order common knowledge observation.
- seed: int, Random seed.
- random_start_player: bool, Random start player.
"""
assert isinstance(config, dict), "Expected config to be of type dict."
self.game = pyhanabi.HanabiGame(config)
self.observation_encoder = pyhanabi.ObservationEncoder(
self.game, pyhanabi.ObservationEncoderType.CANONICAL)
self.players = self.game.num_players()
self.hist = self.init_hist()
def __init__(self, config):
r"""Creates an environment with the given game configuration.
Args:
config: dict, With parameters for the game. Config takes the following
keys and values.
- colors: int, Number of colors \in [2,5].
- ranks: int, Number of ranks \in [2,5].
- players: int, Number of players \in [2,5].
- hand_size: int, Hand size \in [4,5].
- max_information_tokens: int, Number of information tokens (>=0).
- max_life_tokens: int, Number of life tokens (>=1).
- observation_type: int.
0: Minimal observation.
1: First-order common knowledge observation.
- seed: int, Random seed.
- random_start_player: bool, Random start player.
"""
assert isinstance(config, dict), "Expected config to be of type dict."
self.game = pyhanabi.HanabiGame(config)
self.observation_encoder = pyhanabi.ObservationEncoder(
self.game, pyhanabi.ObservationEncoderType.CANONICAL)
self.players = self.game.num_players()
# in case the game_config did not contain specific keys because they were meant to be defaulted
config['hand_size'] = self.game.hand_size()
config['num_players'] = self.game.num_players()
config['num_colors'] = self.game.num_colors()
config['num_ranks'] = self.game.num_ranks()
self.reward_metrics = RewardMetrics(config)
self.augment_input = USE_AUGMENTED_NETWORK_INPUTS_WHEN_WRAPPING_ENV
self.augment_input_using_binary = USE_AUGMENTED_BINARY_INPUTS_WHEN_WRAPPING_ENV
self.observation_augmenter = ObservationAugmenter(config, use_binary=self.augment_input_using_binary)
self.OPEN_HANDS = OPEN_HANDS
def run_game(game_parameters):
"""Play a game, selecting random actions."""
def print_state(state):
"""Print some basic information about the state."""
print("")
print("Current player: {}".format(state.cur_player()))
print(state)
# Example of more queries to provide more about this state. For
# example, bots could use these methods to to get information
# about the state in order to act accordingly.
print("### Information about the state retrieved separately ###")
print("### Information tokens: {}".format(state.information_tokens()))
print("### Life tokens: {}".format(state.life_tokens()))
print("### Fireworks: {}".format(state.fireworks()))
print("### Deck size: {}".format(state.deck_size()))
print("### Discard pile: {}".format(str(state.discard_pile())))
print("### Player hands: {}".format(str(state.player_hands())))
print("")
def print_observation(observation):
"""Print some basic information about an agent observation."""
print("--- Observation ---")
print(observation)
print("### Information about the observation retrieved separately ###")
print("### Current player, relative to self: {}".format(
observation.cur_player_offset()))
print("### Observed hands: {}".format(observation.observed_hands()))
print("### Card knowledge: {}".format(observation.card_knowledge()))
print("### Discard pile: {}".format(observation.discard_pile()))
print("### Fireworks: {}".format(observation.fireworks()))
print("### Deck size: {}".format(observation.deck_size()))
move_string = "### Last moves:"
for move_tuple in observation.last_moves():
move_string += " {}".format(move_tuple)
print(move_string)
print("### Information tokens: {}".format(observation.information_tokens()))
print("### Life tokens: {}".format(observation.life_tokens()))
print("### Legal moves: {}".format(observation.legal_moves()))
print("--- EndObservation ---")
def print_encoded_observations(encoder, state, num_players):
print("--- EncodedObservations ---")
print("Observation encoding shape: {}".format(encoder.shape()))
print("Current actual player: {}".format(state.cur_player()))
for i in range(num_players):
print("Encoded observation for player {}: {}".format(
i, encoder.encode(state.observation(i))))
print("--- EndEncodedObservations ---")
game = pyhanabi.HanabiGame(game_parameters)
print(game.parameter_string(), end="")
obs_encoder = pyhanabi.ObservationEncoder(
game, enc_type=pyhanabi.ObservationEncoderType.CANONICAL)
state = game.new_initial_state()
while not state.is_terminal():
if state.cur_player() == pyhanabi.CHANCE_PLAYER_ID:
state.deal_random_card()
continue
print_state(state)
observation = state.observation(state.cur_player())
print_observation(observation)
print_encoded_observations(obs_encoder, state, game.num_players())
legal_moves = state.legal_moves()
print("")
print("Number of legal moves: {}".format(len(legal_moves)))
move = np.random.choice(legal_moves)
print("Chose random legal move: {}".format(move))
state.apply_move(move)
print("")
print("Game done. Terminal state:")
print("")
print(state)
print("")
print("score: {}".format(state.score()))