def reset(self):
"""Resets the state of the environment and returns an initial observation.
Returns:
observation (object): the initial observation.
"""
logger.info("Reset called")
self.initialize()
return self.observe(phase=0)
def __init__(self, num_players, roles=None, flex=0):
"""
:param num_players: int, number of player, must be grater than 4
:param roles: list of str, list of roles for each agent
:param flex: float [0,1), percentage of targets to consider when voting, 0 is just one, depend on the number of player.
EG: if num_players=10 -> targets are list of 10 elements, 10*0.5=5 -> first 5 player are considered when voting
"""
if isinstance(num_players, EnvContext):
try:
num_players = num_players['num_players']
except KeyError:
raise AttributeError(
f"Attribute 'num_players' should be present in the EnvContext"
)
# number of player should be more than 5
assert num_players >= 5, "Number of player should be >= 5"
if roles is None:
# number of wolves should be less than villagers
num_wolves = math.floor(math.sqrt(num_players))
num_villagers = num_players - num_wolves
roles = [ww] * num_wolves + [vil] * num_villagers
random.shuffle(roles)
logger.info(
f"Starting game with {num_players} players: {num_villagers} {vil} and {num_wolves} {ww}"
)
else:
assert len(
roles
) == num_players, f"Length of role list ({len(roles)}) should be equal to number of players ({num_players})"
self.num_players = num_players
self.roles = roles
self.penalties = CONFIGS['penalties']
self.max_days = CONFIGS['max_days']
if flex == 0:
self.flex = 1
else:
self.flex = math.floor(num_players * flex)
# define empty attributes, refer to initialize method for more info
self.role_map = None
self.status_map = None
self.is_night = True
self.is_comm = True
self.day_count = 0
self.is_done = False
self.targets = None
self.previous_target = None
self.custom_metrics = None
self.initialize()
def reset(self):
"""Resets the state of the environment and returns an initial observation.
Returns:
observation (object): the initial observation.
"""
if Params.log_step == self.ep_step:
logger.info("Reset called")
self.initialize()
init_signal = {p: [-1] * self.signal_length for p in range(self.num_players)}
obs = self.observe(phase=0, signal=init_signal, targets={k: -1 for k in range(self.num_players)})
obs, _, _, _ = self.convert(obs, {}, {}, {}, 0)
return obs
def check_done(self, rewards):
"""
Check if the game is over, moreover return true for dead agent in done
:param rewards: dict, maps agent id_ to curr reward
:return:
dones: list of bool statement
rewards: update rewards
"""
dones = {id_: 0 for id_ in rewards.keys()}
for idx in range(self.num_players):
# done if the player is not alive
done = not self.status_map[idx]
dones[idx] = done
# get list of alive agents
alives = self.get_ids('all', alive=True)
# if there are more wolves than villagers than they won
wolf_won = len(self.get_ids(ww)) >= len(self.get_ids(vil))
# if there are no more wolves than the villager won
village_won = all([
role == vil for id_, role in self.role_map.items() if id_ in alives
])
if wolf_won: # if wolves won
# set flag to true (for reset)
self.is_done = True
# reward
for idx in self.get_ids(ww, alive=False):
rewards[idx] += self.penalties.get('victory')
for idx in self.get_ids(vil, alive=False):
rewards[idx] += self.penalties.get('lost')
logger.info(f"\n{'#' * 10}\nWolves won\n{'#' * 10}\n")
self.custom_metrics['win_wolf'] += 1
if village_won:
self.is_done = True
for idx in self.get_ids(vil, alive=False):
rewards[idx] += self.penalties.get('victory')
for idx in self.get_ids(ww, alive=False):
rewards[idx] += self.penalties.get('lost')
logger.info(f"\n{'#' * 10}\nVillagers won\n{'#' * 10}\n")
self.custom_metrics['win_vil'] += 1
if self.day_count >= self.max_days - 1:
self.is_done = True
return dones, rewards
def __init__(self, configs, roles=None, flex=0):
"""
:param num_players: int, number of player, must be grater than 4
:param roles: list of str, list of roles for each agent
:param flex: float [0,1), percentage of targets to consider when voting, 0 is just one, depend on the number of player.
EG: if num_players=10 -> targets are list of 10 elements, 10*0.5=5 -> first 5 player are considered when voting
"""
# if config is dict
if isinstance(configs, EnvContext) or isinstance(configs, dict):
# get num player
try:
num_players = configs['num_players']
except KeyError:
raise AttributeError(f"Attribute 'num_players' should be present in the EnvContext")
elif isinstance(configs, int):
# used for back compatibility
num_players = configs
else:
raise AttributeError(f"Type {type(configs)} is invalid for config")
# number of player should be more than 5
assert num_players >= 5, "Number of player should be >= 5"
if roles is None:
# number of wolves should be less than villagers
num_wolves = math.floor(math.sqrt(num_players))
num_villagers = num_players - num_wolves
roles = [ww] * num_wolves + [vil] * num_villagers
# random.shuffle(roles)
logger.info(f"Starting game with {num_players} players: {num_villagers} {vil} and {num_wolves} {ww}")
else:
assert len(
roles) == num_players, f"Length of role list ({len(roles)}) should be equal to number of players ({num_players})"
num_wolves = len([elem for elem in roles if elem == ww])
self.num_players = num_players
self.num_wolves = num_wolves
self.roles = roles
self.penalties = CONFIGS['penalties']
self.max_days = CONFIGS['max_days']
self.signal_length = CONFIGS['signal_length']
self.signal_range = CONFIGS['signal_range']
# used for logging game
self.ep_step = 0
if flex == 0:
self.flex = 1
else:
self.flex = math.floor(num_players * flex)
# define empty attributes, refer to initialize method for more info
self.status_map = None
self.shuffle_map = None
self.unshuffle_map = None
self.is_night = True
self.is_comm = True
self.day_count = 0
self.phase = 0
self.is_done = False
self.custom_metrics = None
self.role_map = None
self.initialize()