def update_model(self, experiences):
logger = roomai.get_logger()
reward_plus_gamma_q = []
info_feats = []
action_feats = []
logger = roomai.get_logger()
for experience in experiences:
next_action_feats = [
action_feat
for action_feat in experience.next_available_action_feats
]
next_info_feats = [
experience.next_info_feat
for i in range(len(experience.next_available_action_feats))
]
q = self.sess.run(self.q,
feed_dict={
self.info_feats: next_info_feats,
self.action_feats: next_action_feats
})
reward_plus_gamma_q.append(experience.reward +
self.gamma * np.max(q))
info_feats.append(experience.info_feat)
action_feats.append(experience.action_feat)
_, loss, q = self.sess.run(
(self.train_op, self.loss, self.q),
feed_dict={
self.info_feats: info_feats,
self.action_feats: action_feats,
self.reward_plus_gamma_q: reward_plus_gamma_q
})
def compete_silent(cls, env, players):
'''
Use the game environment to hold a compete_silent for the players
:param env: The game environment
:param players: The normal players (without the chance player)
:return: scores for the players
'''
total_scores = [0 for i in range(len(players))]
total_count = 10
num_normal_players = len(players)
players = players + [roomai.games.common.RandomPlayerChance()]
for count in range(total_count):
chips = [(1000 + int(random.random() * 200))
for i in range(len(players))]
dealer_id = int(random.random() * len(players))
big_blind_bet = 50
infos, public, persons, private, action_history = env.init({
"chips":
chips,
"param_num_normal_players":
num_normal_players,
"dealer_id":
dealer_id,
"big_blind_bet":
big_blind_bet
})
for i in range(len(players)):
players[i].receive_info(infos[i])
while public[-1].is_terminal == False:
turn = public[-1].turn
action = players[turn].take_action()
# print len(infos[turn].person_state.available_actions),action.key(),turn
infos, public, persons, private, _ = env.forward(action)
for i in range(len(players)):
players[i].receive_info(infos[i])
for i in range(len(players)):
players[i].receive_info(infos[i])
for i in range(num_normal_players):
total_scores[i] += public[-1].scores[i]
if (count + 1) % 500 == 0:
tmp_scores = [0 for i in range(len(total_scores))]
for i in range(len(total_scores)):
tmp_scores[i] = total_scores[i] / (count + 1)
roomai.get_logger().info(
"TexasHoldem completes %d competitions, scores=%s" %
(count + 1, ",".join([str(i) for i in tmp_scores])))
for i in range(len(total_scores)):
total_scores[i] /= 1.0 * total_count
return total_scores
def compete(cls, env, players):
"""
Args:
env:
players:
Returns:
"""
total_scores = [0 for i in xrange(len(players))]
total_count = 1000
for count in range(total_count):
chips = [(1000 + int(random.random() * 200))
for i in range(len(players))]
num_players = len(players)
dealer_id = int(random.random() * len(players))
big_blind_bet = 50
infos, public, persons, private = env.init({
"chips":
chips,
"num_players":
num_players,
"dealer_id":
dealer_id,
"big_blind_bet":
big_blind_bet
})
for i in xrange(len(players)):
players[i].receive_info(infos[i])
while public.is_terminal == False:
turn = public.turn
action = players[turn].take_action()
#print len(infos[turn].person_state.available_actions),action.key(),turn
infos, public, persons, private = env.forward(action)
for i in xrange(len(players)):
players[i].receive_info(infos[i])
for i in xrange(len(players)):
players[i].receive_info(infos[i])
total_scores[i] += public.scores[i]
if (count + 1) % 500 == 0:
tmp_scores = [0 for i in xrange(len(total_scores))]
for i in xrange(len(total_scores)):
tmp_scores[i] = total_scores[i] / (count + 1)
roomai.get_logger().info(
"TexasHoldem completes %d competitions, scores=%s" %
(count + 1, ",".join([str(i) for i in tmp_scores])))
for i in xrange(len(total_scores)):
total_scores[i] /= 1.0 * total_count
return total_scores
def init(self, params=dict()):
'''
Initialize the TexasHoldem game environment with the initialization params.\n
The initialization is a dict with only an option: \n
param_num_normal_players: how many players are in the game, the option must be in {2, 4, 5}, default 5. An example of the initialization param is {"param_num_normal_players":2} \n
:param params: the initialization params
:return: infos, public_state_history, person_states_history, private_state_history, playerid_action_history
'''
logger = roomai.get_logger()
############ public state and private state ##########
public_state = BangStatePublic()
private_state = BangStatePrivate()
self.__public_state_history__.append(public_state)
self.__private_state_history__.append(private_state)
if "param_num_normal_players" in params:
public_state.__param_num_normal_players__ = params[
"param_num_normal_players"]
else:
public_state.__param_num_normal_players__ = 5
if public_state.param_num_normal_players not in [2, 4, 5]:
logger.fatal("The number of normal players must be in [2,4,5]")
raise ValueError("The number of normal players must be in [2,4,5]")
public_state.__public_player_infos__ = [
PublicPlayerInfo()
for i in range(public_state.__param_num_normal_players__)
]
for i in range(public_state.__param_num_normal_players__):
public_state.__public_player_infos__[i].__num_hand_cards__ = 0
public_state.__public_player_infos__[i].__character_card__ = None
public_state.__public_player_infos__[i].__equipment_cards__ = []
public_state.__phase_info__ = PhaseInfo()
public_state.__phase_info__.__playid__ = public_state.__param_num_normal_players__
public_state.__phase_info__.__phase__ = PhaseInfo.ChancePlay
public_state.__turn__ = public_state.__param_num_normal_players__
########### person states #########
person_states = [
BangStatePerson()
for i in range(public_state.param_num_normal_players + 1)
]
for i in range(public_state.param_num_normal_players):
self.__person_states_history__[i].append(person_states[i])
person_states[i][0].__id__ = i
person_states[i][0].__hand_cards__ = []
person_states[i][0].__role__ = None
person_states[i][0].__available_actions__ = dict()
person_states[public_state.__param_num_normal_players__][
0].__available_actions__ = self.available_actions()
return self.__gen_infos__(
), self.__public_state_history__, self.__person_states_history__, self.__private_state_history__, self.__playerid_action_history__
def sampling(probs):
logger = roomai.get_logger()
r = random.random()
sum1 = 0
for i in range(len(probs)):
sum1 += probs[i]
if sum1 > r:
return i
logger.warn("Sampling probs(%s) with r = %f occurs sum(probs) >= r", ",".join([str(i) for i in probs]), r)
return len(probs)-1
def lookup(cls, key):
'''
lookup a PokerCard with the specified key
:param key: The specified key
:return: The PokerCard with the specified key
'''
logger = roomai.get_logger()
if key not in AllPokerCards:
logger.fatal("key (%s) is not invalid poker card key"%(key))
raise ValueError("key (%s) is not invalid poker card key"%(key))
return AllPokerCards[key]
def gen_info_feat(self, info):
logger = roomai.get_logger()
hand_cards = info.person_state.hand_cards
info_feat = np.zeros((self.num_point, self.num_suit, self.info_dim))
current_id = info.person_state.id
previous_id = (current_id + 3 - 1) % 3
next_id = (current_id + 1) % 3
if info.public_state.stage == 0:
for card in hand_cards:
info_feat[card.point_rank, card.suit_rank, 0] += 1
for person_action in info.public_state.action_history:
person_id = person_action[0]
action = person_action[1]
for card in action.cards:
if person_id == current_id:
info_feat[card.point_rank, card.suit_rank, 1] += 1
elif person_id == previous_id:
info_feat[card.point_rank, card.suit_rank, 2] += 1
elif person_id == next_id:
info_feat[card.point_rank, card.suit_rank, 3] += 1
elif person_id == 3:
logger.debug(
"SevenKingModel finds the chance player-action pair in public_state.action_history"
)
else:
for card in hand_cards:
info_feat[card.point_rank, card.suit_rank, 4] += 1
for person_action in info.public_state.action_history:
person_id = person_action[0]
action = person_action[1]
for card in action.cards:
if person_id == current_id:
info_feat[card.point_rank, card.suit_rank, 5] += 1
elif person_id == previous_id:
info_feat[card.point_rank, card.suit_rank, 6] += 1
elif person_id == next_id:
info_feat[card.point_rank, card.suit_rank, 7] += 1
elif person_id == 3:
logger.debug(
"SevenKingModel finds the chance player-action pair in public_state.action_history"
)
return info_feat
def init(self, params = dict()):
'''
Initialize the TexasHoldem game environment with the initialization params.\n
The initialization is a dict with some options\n
1. param_num_normal_players: how many players are in the game, the option must be in {2, 4, 5}, default 5\n
2. param_start_turn: The param_start_turn is the id of a normal player, who is the first to take an action \n
An example of the initialization param is {"param_num_normal_players":2}
:param params: the initialization params
:return: infos, public_state, person_states, private_state
'''
logger = roomai.get_logger()
public_state = BangStatePublic()
private_state = BangStatePrivate()
if "param_num_normal_players" in params:
public_state.__param_num_normal_players__ = params["param_num_normal_players"]
else:
public_state.__param_num_normal_players__ = 5
if public_state.param_num_normal_players not in [2,4,5]:
logger.fatal("The number of normal players must be in [2,4,5]")
raise ValueError("The number of normal players must be in [2,4,5]")
public_state.__public_person_info__ = [PublicPersonInfo() for i in range(public_state.__param_num_normal_players__)]
for i in range(public_state.__param_num_normal_players__):
public_state.__public_person_info__[i].__num_hand_cards__ = 0
public_state.__public_person_info__[i].__charactor_card__ = None
public_state.__public_person_info__[i].__equipment_cards__ = []
person_states = [BangStatePerson() for i in range(public_state.param_num_normal_players+1)]
self.__public_state_history__.append(public_state)
self.__private_state_history__.append(private_state)
for i in range(public_state.param_num_normal_players):
self.__person_states_history__[i].append(person_states[i])
self.__person_states_history__[i][0].__id__ = i
self.__person_states_history__[i][0].__hand_cards__ = []
self.__person_states_history__[i][0].__role__ = ""
self.__person_states_history__[public_state.__param_num_normal_players__][0].__available_actions__ = self.available_actions()
self.__gen_infos__()
def lookup(cls, key):
'''
lookup a Card with the specified key
:param key: The specified key
:return: The Card with the specified key
'''
logger = roomai.get_logger()
if key not in AllPlayingCardsDict:
logger.fatal("key (%s) is not invalid poker normalcard key" %
(key))
raise ValueError("key (%s) is not invalid poker normalcard key" %
(key))
return AllPlayingCardsDict[key]
def lookup(self, key):
logger = roomai.get_logger()
if key is None or not isinstance(key, str):
logger.fatal(
"In the constructor BangActionChance.lookup(key), the key must be a str"
)
raise TypeError(
"In the constructor BangActionChance.lookup(key), the key must be a str"
)
if key not in AllBangActionChancesDict:
logger.fatal(
"In the constructor BangActionChance.lookup(key), the key must be the key of CharacterCard, RoleCard or PlayingCard"
)
raise ValueError(
"In the constructor BangActionChance.lookup(key), the key must be the key of CharacterCard, RoleCard or PlayingCard"
)
return AllBangActionChancesDict[key]
def __init__(self, key):
logger = roomai.get_logger()
self.__type__ = BangActionType.card
self.__key__ = None
self.__skill__ = None
self.__seen_cards__ = set()
self.__choosen_cards__ = set()
self.__card__ = None
self.__card_targets__ = []
self.__other__ = None
self.__other_targets__ = []
keys = key.split("_")
if keys[0] in AllPlayingCardsDict:
self.__card__ = AllPlayingCardsDict[keys[0]]
else:
logger.info("%s is invalid action key, since the cardkey %s is invalid"%(key, keys[0]))
def __action_chance__(self, action):
pu = self.__public_state_history__[-1]
pes = [self.__person_states_history__[i][-1] for i in range(len(self.__person_states_history__))]
pr = self.__private_state_history__[-1]
card = action.card
num = len(pr.all_used_cards)
if num < (len(pes)-1) * 2:
idx = int(num / 2)
pes[idx].__hand_cards__.append(card)
elif num < (len(pes)-1) * 2 + 5:
pr.__keep_cards__.append(card)
else:
logger = roomai.get_logger()
logger.fatal("the chance action in the invalid condition")
pr.__all_used_cards__.append(card)
def __init__(self, key):
logger = roomai.get_logger()
self.__is_public__ = False
if key in AllPlayingCardsDict:
self.__type__ = BangActionChanceType.playingcard
self.__card__ = PlayingCard.lookup(key)
self.__key__ = key
elif key in AllCharacterCardsDict:
self.__type__ = BangActionChanceType.charactercard
self.__card__ = CharacterCard.lookup(key)
self.__key__ = key
elif key in AllRoleCardsDict:
self.__type__ = BangActionChanceType.rolecard
self.__card__ = RoleCard.lookup(key)
self.__key__ = key
else:
logger.fatal(
"In the constructor BangActionChance(card), the parameter card must be NormalCard, CharacterCard or RoleCard"
)
raise TypeError(
"In the constructor BangActionChance(card), the parameter card must be NormalCard, CharacterCard or RoleCard"
)
def __init__(self, card):
logger = roomai.get_logger()
self.__is_public__ = False
if isinstance(card, PlayingCard):
self.__type__ = BangActionChanceType.playingcard
self.__card__ = card
self.__key__ = card.key
elif isinstance(card, CharacterCard):
self.__type__ = BangActionChanceType.charactercard
self.__card__ = card
self.__key__ = card.key
elif isinstance(card, RoleCard):
self.__type__ = BangActionChanceType.rolecard
self.__card__ = card
self.__key__ = card.key
else:
logger.fatal(
"In the constructor BangActionChance(card), the parameter card must be NormalCard, CharacterCard or RoleCard"
)
raise TypeError(
"In the constructor BangActionChance(card), the parameter card must be NormalCard, CharacterCard or RoleCard"
)
def __init__(self, role):
logger = roomai.get_logger()
if isinstance(role, str):
logger.fatal(
"In the constructor RoleCard(rolecard), the rolecard must be a str."
)
raise TypeError(
"In the constructor RoleCard(rolecard), the rolecard must be a str."
)
if role not in [
RoleCardNames.sheriff, RoleCardNames.deputy_sheriff,
RoleCardNames.outlaw, RoleCardNames.renegade
]:
logger.fatal(
"In the constructor RoleCard(rolecard), the rolecard must be one of [%s,%s,%s,%s]"
% (RoleCardNames.sheriff, RoleCardNames.deputy_sheriff,
RoleCardNames.outlaw, RoleCardNames.renegade))
raise TypeError(
"In the constructor RoleCard(rolecard), the rolecard must be one of [%s,%s,%s,%s]"
% (RoleCardNames.sheriff, RoleCardNames.deputy_sheriff,
RoleCardNames.outlaw, RoleCardNames.renegade))
self.__name1__ = role
#!/bin/python
#coding=utf8
import roomai
import roomai.common
logger = roomai.get_logger()
######################################################################### Basic Concepts #####################################################
class AbstractPublicState(object):
'''
The abstract class of the public state. The information in the public state is public to every player
'''
def __init__(self):
self.__turn__ = None
self.__action_history__ = []
self.__is_terminal__ = False
self.__scores__ = None
def __get_turn__(self): return self.__turn__
turn = property(__get_turn__, doc = "The players[turn] is expected to take an action.")
def __get_action_history__(self): return tuple(self.__action_history__)
action_history = property(__get_action_history__, doc = "The action_history so far. For example, action_history = [(0, roomai.kuhn.KuhnAction.lookup(\"check\"),(1,roomai.kuhn.KuhnAction.lookup(\"bet\")]")
'''
def __get_previous_id__(self): return self.__previous_id__
previous_id = property(__get_previous_id__,doc = "The players[previous_id] took an action in the previous epoch. In the first epoch, previous_id is None")
def lookup(cls, key):
logger = roomai.get_logger()
if key not in AllRoleCardsDict:
logger.fatal("%s is not valid rolecard key" % (key))
raise TypeError("%s is not valid rolecard key" % (key))
return AllRoleCardsDict[key]
def init(self, params=dict()):
'''
Initialize the TexasHoldem game environment with the initialization params.\n
The initialization is a dict with some options\n
1) allcards: the order of all poker cards appearing\n
2) record_history: whether to record all history states. if you need call the backward function, please set it to True. default False\n
3) num_players: how many players are in the game, default 3\n
4) dealer_id: the player id of the dealer, default random\n
5) chips: the initialization chips, default [1000,1000,...]\n
6) big_blind_bet: the number of chips for the big blind bet, default 10\n
An example of the initialization param is {"num_players":2,"record_history":True}
:param params: the initialization params
:return: infos, public_state, person_states, private_state
'''
self.logger = roomai.get_logger()
if "num_players" in params:
self.__params__["num_players"] = params["num_players"]
else:
self.__params__["num_players"] = 3
if "dealer_id" in params:
self.__params__["dealer_id"] = params["dealer_id"]
else:
self.__params__["dealer_id"] = int(random.random() *
self.__params__["num_players"])
if "chips" in params:
self.__params__["chips"] = params["chips"]
else:
self.__params__["chips"] = [
1000 for i in range(self.__params__["num_players"])
]
if "big_blind_bet" in params:
self.__params__["big_blind_bet"] = params["big_blind_bet"]
else:
self.__params__["big_blind_bet"] = 10
if "allcards" in params:
self.__params__["allcards"] = [
c.__deepcopy__() for c in params["allcards"]
]
else:
self.__params__["allcards"] = list(
roomai.common.AllPokerCards_Without_King.values())
random.shuffle(self.__params__["allcards"])
if "record_history" in params:
self.__params__["record_history"] = params["record_history"]
else:
self.__params__["record_history"] = False
self.__check_initialization_configuration__(self)
## public info
small = (self.__params__["dealer_id"] +
1) % self.__params__["num_players"]
big = (self.__params__["dealer_id"] +
2) % self.__params__["num_players"]
self.public_state = TexasHoldemPublicState()
pu = self.public_state
pu.__num_players__ = self.__params__["num_players"]
pu.__dealer_id__ = self.__params__["dealer_id"]
pu.__big_blind_bet__ = self.__params__["big_blind_bet"]
pu.__raise_account__ = self.__params__["big_blind_bet"]
pu.__is_fold__ = [False for i in range(self.__params__["num_players"])]
pu.__num_fold__ = 0
pu.__is_allin__ = [
False for i in range(self.__params__["num_players"])
]
pu.__num_allin__ = 0
pu.__is_needed_to_action__ = [
True for i in range(self.__params__["num_players"])
]
pu.__num_needed_to_action__ = pu.num_players
pu.__bets__ = [0 for i in range(self.__params__["num_players"])]
pu.__chips__ = self.__params__["chips"]
pu.__stage__ = StageSpace.firstStage
pu.__turn__ = (big + 1) % pu.num_players
pu.__public_cards__ = []
pu.__previous_id__ = None
pu.__previous_action__ = None
if pu.chips[big] > self.__params__["big_blind_bet"]:
pu.__chips__[big] -= self.__params__["big_blind_bet"]
pu.__bets__[big] += self.__params__["big_blind_bet"]
else:
pu.__bets__[big] = pu.chips[big]
pu.__chips__[big] = 0
pu.__is_allin__[big] = True
pu.__num_allin__ += 1
pu.__max_bet_sofar__ = pu.bets[big]
pu.__raise_account__ = self.__params__["big_blind_bet"]
if pu.chips[small] > self.__params__["big_blind_bet"] / 2:
pu.__chips__[small] -= self.__params__["big_blind_bet"] / 2
pu.__bets__[small] += self.__params__["big_blind_bet"] / 2
else:
pu.__bets__[small] = pu.chips[small]
pu.__chips__[small] = 0
pu.__is_allin__[small] = True
pu.__num_allin__ += 1
pu.__is_terminal__ = False
pu.__scores__ = [0 for i in range(self.__params__["num_players"])]
# private info
self.private_state = TexasHoldemPrivateState()
pr = self.private_state
pr.__keep_cards__ = self.__params__[
"allcards"][self.__params__["num_players"] *
2:self.__params__["num_players"] * 2 + 5]
## person info
self.person_states = [
TexasHoldemPersonState()
for i in range(self.__params__["num_players"])
]
pes = self.person_states
for i in range(self.__params__["num_players"]):
pes[i].__id__ = i
pes[i].__hand_cards__ = self.__params__["allcards"][i * 2:(i + 1) *
2]
pes[pu.turn].__available_actions__ = self.available_actions(
pu, pes[pu.turn])
self.__gen_history__()
infos = self.__gen_infos__()
if self.logger.level <= logging.DEBUG:
self.logger.debug("TexasHoldemEnv.init: num_players = %d, dealer_id = %d, chip = %d, big_blind_bet = %d"%(\
pu.num_players,\
pu.dealer_id,\
pu.chips[0],\
pu.big_blind_bet
))
return infos, pu, pes, pr
def available_actions(self):
'''
Generate all valid actions given the public state and the person state
:return: all valid actions
'''
logger = roomai.get_logger()
###################################### chance action #################################
## charactercard
if self.__public_state_history__[-1].__public_person_infos__[-1].__character_card__ is None:
available_actions = dict()
tmp_set = set()
for i in range(len(self.__public_state_history__[-1].__public_person_infos__)):
if self.__public_state_history__[-1].__public_person_infos__[i].__character_card__ is not None:
tmp_set.add(self.__public_state_history__[-1].__public_person_infos__[i].__character_card__.key)
for key in AllCharacterCardsDict:
if key not in tmp_set:
available_actions[key] = BangActionChance.lookup(key)
return available_actions
## rolecard
for i in range(self.__public_state_history__[-1].param_num_normal_players):
if self.__person_states_history__[i].person_states[-1].__role_card__ is None:
available_actions = dict()
tmp_set = set()
num_sheriff = 0
num_deputy_sheriff = 0
num_renegade = 0
num_outlaw = 0
for j in range(len(self.__public_state_history__[-1].param_num_normal_players)):
if self.__person_states_history__[j].person_states[-1].__role_card__ is not None:
tmp_set.add(self.__person_states_history__[j].person_states[-1].__role_card__.key)
if self.__public_state_history__[-1].__param_num_normal_players__ == 2:
return available_actions
elif self.__public_state_history__[-1].__param_num_normal_players__ == 4:
num_sheriff = 1
num_renegade = 1
num_outlaw = 2
elif self.__public_state_history__[-1].__param_num_normal_players__ == 5:
num_sheriff = 1
num_deputy_sheriff = 1
num_renegade = 1
num_outlaw = 2
else:
logger.fatal("param_num_normal_players not in [2,4,5]")
raise ValueError("param_num_normal_players not in [2,4,5]")
for key in tmp_set:
if key == CardRole.RoleCardNames.sheriff:
num_sheriff = num_sheriff - 1
if key == CardRole.RoleCardNames.deputy_sheriff:
num_deputy_sheriff = num_deputy_sheriff - 1
if key == CardRole.RoleCardNames.renegade:
num_renegade = num_renegade - 1
if key == CardRole.RoleCardNames.outlaw:
num_outlaw = num_outlaw - 1
if num_sheriff > 0:
available_actions[CardRole.RoleCardNames.sheriff] = BangActionChance.lookup(CardRole.RoleCardNames.sheriff)
if num_deputy_sheriff > 0:
available_actions[CardRole.RoleCardNames.deputy_sheriff] = BangActionChance.lookup(CardRole.RoleCardNames.deputy_sheriff)
if num_renegade > 0:
available_actions[CardRole.RoleCardNames.renegade] = BangActionChance.lookup(CardRole.RoleCardNames.renegade)
if num_outlaw > 0:
available_actions[CardRole.RoleCardNames.outlaw] = BangActionChance.lookup(CardRole.RoleCardNames.outlaw)
return available_actions
## deal cards
available_actions = dict()
for card in self.__private_state_history__.deck:
available_actions[card.key] = card
####################################### action ####################################
turn = self.__public_state_history__[-1].turn
tmp_set = dict()
if len(self.__public_state_history__[-1].response_infos_stack) > 0:
person_states = self.__person_states_history__[-1]
private_state = self.__private_state_history__[-1]
subject = self.__public_state_history__[-1].response_infos_stack[-1].subject
object = self.__public_state_history__[-1].response_infos_stack[-1].object
reason = self.__public_state_history__[-1].response_infos_stack[-1].reason
if reason == ResponseInfo.UseIndian:
for card in person_states[subject].hand_cards:
if card.name == PlayingCardNames.Bang:
tmp_set[card.name] = BangAction.lookup(card.name)
tmp_set[OtherActionNames.giveup] = BangAction.lookup(OtherActionNames.giveup)
return tmp_set
elif reason == ResponseInfo.UseCatling:
for card in person_states[subject].hand_cards:
if card.name == PlayingCardNames.Miss:
tmp_set[card.name] = BangAction.lookup(card.name)
tmp_set[OtherActionNames.giveup] = BangAction.lookup(OtherActionNames.giveup)
return tmp_set
elif reason == ResponseInfo.ToDead:
for card in person_states[subject].hand_cards:
if card.name == PlayingCardNames.Beer:
tmp_set[card.name+"_%d"%(object)] = BangAction.lookup(card.name+"_%d"%(object))
tmp_set[OtherActionNames.giveup] = BangAction.lookup(OtherActionNames.giveup)
return tmp_set
elif reason == ResponseInfo.Shuffle:
for card in private_state.shuffle_deck:
tmp_set[card.key] = BangActionChance.lookup(card.key)
return tmp_set
def init(self, params=dict()):
'''
Initialize the TexasHoldem game environment with the initialization params.\n
The initialization is a dict with some options\n
1. param_num_normal_players: how many players are in the game, default 3\n
2. param_dealer_id: the player id of the dealer, default random\n
3. param_init_chips: the initialization chips, default [1000,1000,...]\n
4. param_big_blind_bet: the number of chips for the big blind bet, default 10\n
An example of the initialization param is {"param_num_normal_players":2,"backward_enable":True}
:param params: the initialization params
:return: infos, public_state, person_states, private_state
'''
logger = roomai.get_logger()
public_state = TexasHoldemStatePublic()
self.__public_state_history__.append(public_state)
if "param_num_normal_players" in params:
public_state.__param_num_normal_players__ = params[
"param_num_normal_players"]
else:
public_state.__param_num_normal_players__ = 3
if "param_start_turn" in params:
public_state.__param_start_turn__ = params["param_start_turn"]
else:
public_state.__param_start_turn__ = int(
random.random() * public_state.param_num_normal_players)
if "param_dealer_id" in params:
public_state.__param_dealer_id__ = params["param_dealer_id"]
else:
public_state.__param_dealer_id__ = int(
random.random() * public_state.param_num_normal_players)
if "param_init_chips" in params:
public_state.__param_init_chips__ = params["param_init_chips"]
else:
public_state.__param_init_chips__ = [
1000 for i in range(public_state.param_num_normal_players)
]
if "param_big_blind_bet" in params:
public_state.__param_big_blind_bet__ = params[
"param_big_blind_bet"]
else:
public_state.__param_big_blind_bet__ = 10
## check initialization config
if len(public_state.param_init_chips
) != public_state.param_num_normal_players:
raise ValueError(
"len(env.param_initialization_chips) %d != param_num_normal_players %d"
% (len(public_state.param_init_chips),
public_state.num_normal_players))
if public_state.param_num_normal_players > 6:
raise ValueError(
"The maximum of the number of players is 6. Now, the number of players = %d"
% (public_state.param_num_normal_players))
## public info
small = (public_state.param_dealer_id +
1) % public_state.param_num_normal_players
big = (public_state.param_dealer_id +
2) % public_state.param_num_normal_players
pu = public_state
pu.__is_fold__ = [
False for i in range(public_state.param_num_normal_players)
]
pu.__num_fold__ = 0
pu.__is_allin__ = [
False for i in range(public_state.param_num_normal_players)
]
pu.__num_allin__ = 0
pu.__is_needed_to_action__ = [
True for i in range(public_state.param_num_normal_players)
]
pu.__num_needed_to_action__ = pu.param_num_normal_players
pu.__bets__ = [0 for i in range(public_state.param_num_normal_players)]
pu.__chips__ = list(public_state.param_init_chips)
pu.__stage__ = Stage.firstStage
pu.__turn__ = pu.param_num_normal_players
pu.__public_cards__ = []
pu.__previous_id__ = None
pu.__previous_action__ = None
if pu.chips[big] > public_state.param_big_blind_bet:
pu.__chips__[big] -= public_state.param_big_blind_bet
pu.__bets__[big] += public_state.param_big_blind_bet
else:
pu.__bets__[big] = pu.chips[big]
pu.__chips__[big] = 0
pu.__is_allin__[big] = True
pu.__num_allin__ += 1
pu.__max_bet_sofar__ = pu.bets[big]
pu.__raise_account__ = public_state.param_big_blind_bet
if pu.chips[small] > public_state.param_big_blind_bet / 2:
pu.__chips__[small] -= public_state.param_big_blind_bet / 2
pu.__bets__[small] += public_state.param_big_blind_bet / 2
else:
pu.__bets__[small] = pu.chips[small]
pu.__chips__[small] = 0
pu.__is_allin__[small] = True
pu.__num_allin__ += 1
pu.__is_terminal__ = False
pu.__scores__ = [
0 for i in range(public_state.param_num_normal_players)
]
# private info
pr = TexasHoldemStatePrivate()
self.__private_state_history__.append(pr)
pr.__keep_cards__ = []
##pr.__keep_cards__ =allcards[public_state.param_num_normal_players*2:public_state.param_num_normal_players*2+5]
## person info
self.__person_states_history__ = [
[] for i in range(pu.param_num_normal_players + 1)
]
for i in range(pu.param_num_normal_players + 1):
self.__person_states_history__[i].append(TexasHoldemStatePerson())
self.__person_states_history__[i][0].__id__ = i
self.__person_states_history__[i][0].__hand_cards__ = []
self.__person_states_history__[
pu.turn][0].__available_actions__ = self.available_actions()
infos = self.__gen_infos__()
if logger.level <= logging.DEBUG:
logger.debug("TexasHoldemEnv.init: param_num_normal_players = %d, param_dealer_id = %d, param_initialization_chip = %d, param_big_blind_bet = %d"%(\
pu.param_num_normal_players,\
pu.param_dealer_id,\
pu.param_init_chips[0],\
pu.param_big_blind_bet
))
return infos, self.__public_state_history__, self.__person_states_history__, self.__private_state_history__, self.__playerid_action_history__
def forward(self, action):
'''
The TexasHoldem game environments steps with the action taken by the current player
:param action: The action taken by the current player
:return: infos, public_state, person_states, private_state
'''
logger = roomai.get_logger()
pu = self.__public_state_history__[-1].__deepcopy__()
pes = [
self.__person_states_history__[i][-1].__deepcopy__()
for i in range(len(self.__person_states_history__))
]
pr = self.__private_state_history__[-1].__deepcopy__()
self.__public_state_history__.append(pu)
for i in range(len(pes)):
self.__person_states_history__[i].append(pes[i])
self.__private_state_history__.append(pr)
if action.key not in pes[pu.turn].available_actions:
logger.critical("action=%s is invalid" % (action.key))
raise ValueError("action=%s is invalid" % (action.key))
pes[pu.turn].__available_actions__ = dict()
self.__playerid_action_history__.append(
roomai.games.common.ActionRecord(pu.turn, action))
if isinstance(action, TexasHoldemActionChance) == True:
self.__action_chance__(action)
if len(pr.all_used_cards) == (len(pes) - 1) * 2 + 5:
pu.__turn__ = (pu.param_dealer_id + 2 +
1) % pu.param_num_normal_players
pes[pu.turn].__available_actions__ = self.available_actions()
infos = self.__gen_infos__()
return infos, self.__public_state_history__, self.__person_states_history__, self.__private_state_history__, self.__playerid_action_history__
if action.option == TexasHoldemAction.Fold:
self.__action_fold__(action)
elif action.option == TexasHoldemAction.Check:
self.__action_check__(action)
elif action.option == TexasHoldemAction.Call:
self.__action_call__(action)
elif action.option == TexasHoldemAction.Raise:
self.__action_raise__(action)
elif action.option == TexasHoldemAction.AllIn:
self.__action_allin__(action)
else:
raise Exception(
"action.option(%s) not in [Fold, Check, Call, Raise, AllIn]" %
(action.option))
pu.__previous_id__ = pu.turn
pu.__previous_action__ = action
pu.__is_terminal__ = False
pu.__scores__ = [
0 for i in range(
self.__public_state_history__[-1].param_num_normal_players)
]
# computing_score
if TexasHoldemEnv.__is_compute_scores__(
self.__public_state_history__[-1]):
## need showdown
pu.__public_cards__ = pr.keep_cards[0:5]
pu.__is_terminal__ = True
pu.__scores__ = self.__compute_scores__()
# enter into the next stage
elif TexasHoldemEnv.__is_nextround__(
self.__public_state_history__[-1]):
add_cards = []
if pu.stage == Stage.firstStage: add_cards = pr.keep_cards[0:3]
if pu.stage == Stage.secondStage: add_cards = [pr.keep_cards[3]]
if pu.stage == Stage.thirdStage: add_cards = [pr.keep_cards[4]]
pu.__public_cards__.extend(add_cards)
pu.__stage__ = pu.stage + 1
pu.__num_needed_to_action__ = 0
pu.__is_needed_to_action__ = [
False for i in range(pu.param_num_normal_players)
]
for i in range(pu.param_num_normal_players):
if pu.__is_fold__[i] != True and pu.__is_allin__[i] != True:
pu.__is_needed_to_action__[i] = True
pu.__num_needed_to_action__ += 1
pu.__turn__ = pu.param_dealer_id
pu.__turn__ = self.__next_player__(pu)
pes[self.__public_state_history__[-1].
turn].__available_actions__ = self.available_actions()
##normal
else:
pu.__turn__ = self.__next_player__(pu)
self.__person_states_history__[self.__public_state_history__[
-1].turn][-1].__available_actions__ = self.available_actions()
logger = roomai.get_logger()
if logger.level <= logging.DEBUG:
logger.debug("TexasHoldemEnv.forward: num_fold+num_allin = %d+%d = %d, action = %s, stage = %d"%(\
self.__public_state_history__[-1].num_fold,\
self.__public_state_history__[-1].num_allin,\
self.__public_state_history__[-1].num_fold + self.__public_state_history__[-1].num_allin,\
action.key,\
self.__public_state_history__[-1].stage\
))
infos = self.__gen_infos__()
return infos, self.__public_state_history__, self.__person_states_history__, self.__private_state_history__, self.__playerid_action_history__
def forward(self, action):
"""
接受一个动作，先检查是否是ActionChance，处理，如果是正常玩家的，pass
如果是ActionChance，进行相应的动作处理
The Bang game environment steps with the action taken by the current player
:param action
:returns:infos, public_state_history, person_states_history, private_state_history, playerid_action_history
"""
logger = roomai.get_logger()
private_state = copy.deepcopy(self.__private_state_history__[-1])
public_state = copy.deepcopy(self.__public_state_history__[-1])
person_states = [copy.deepcopy(self.__person_states_history__[i][-1]) for i in range(public_state.param_num_normal_players)]
person_states[public_state.turn].__available_actions__ = dict()
self.__public_state_history__.append(public_state)
self.__private_state_history__.append(private_state)
for i in range(len(person_states)):
self.__person_states_history__[i].append(person_states[i])
if isinstance(action, BangActionChance) == True:
if action.type == BangActionChance.BangActionChanceType.charactercard: # chance player deals character cards
person_states[public_state.turn].__available_actions__ = self.available_actions()
for i in range(len(public_state.__public_person_infos__)):
if public_state.__public_person_infos__[i].__character_card is None: # sample a character card to that player
public_state.__public_person_infos__[i].__character_card = \
person_states[public_state.turn].__available_actions__[choice(person_states[public_state.turn].__available_actions__.keys)]
public_state.__turn__ = (public_state.turn - 1) % public_state.param_num_normal_players
return self.__gen_infos__(), self.__public_state_history__, self.__person_states_history__, self.__private_state_history__
# if all players have been assigned a character, return
public_state.__turn__ = (public_state.turn - 1) % public_state.param_num_normal_players
return self.__gen_infos__(), self.__public_state_history__, self.__person_states_history__, self.__private_state_history__
if action.type == BangActionChance.BangActionChanceType.rolecard: # chance player deals role cards
person_states[public_state.turn].__available_actions__ = self.available_actions()
for i in range(public_state.param_num_normal_players):
if person_states[i].__role__ is None: # sample a role card to that player
person_states[i].__role__ = person_states[public_state.turn].__available_actions__[choice(person_states[public_state.turn].__available_actions__.keys)]
public_state.__turn__ = (public_state.turn - 1) % public_state.param_num_normal_players
if person_states[i].__role__ == CardRole.RoleCard(CardRole.RoleCardNames.sheriff):
public_state.__sheriff_id__ = i
return self.__gen_infos__(), self.__public_state_history__, self.__person_states_history__, self.__private_state_history__
public_state.__turn__ = (public_state.turn - 1) % public_state.param_num_normal_players
return self.__gen_infos__(), self.__public_state_history__, self.__person_states_history__, self.__private_state_history__
if action.type == BangActionChance.BangActionChanceType.playingcard: # chance player deals/shuffles cards
person_states[public_state.turn].__available_actions__ = self.available_actions()
private_state.__deal_cards__.append(action)
private_state.__deck__.pop(action.key)
if len(private_state.__deck__) == 0:
# there is no card, then the chance player needs to shuffle discard cards
private_state.__deck__ = public_state.__discard_pile__[:]
public_state.__discard_pile__ = []
public_state.__turn__ = (public_state.turn - 1) % public_state.param_num_normal_players
return self.__gen_infos__(), self.__public_state_history__, self.__person_states_history__, self.__private_state_history__
else:
if len(self.__public_state_history__[-1].response_infos_stack) > 0:
response_action = self.__public_state_history__[-1].response_infos_stack[-1].action
if isinstance(response_action,BangAction) == True \
and response_action.type == BangActionType.card \
and response_action.card.name == PlayingCardNames.Indian:
if action.type == BangActionType.other and action.other == OtherActionNames.giveup:
person_states[public_state.__turn__].__hp__ -= 1
elif action.type == BangActionType.card and action.card.name == PlayingCardNames.Bang:
person_states.__hand_cards__.remove(action.card)
new_turn = (public_state.turn + 1) % (public_state.param_num_normal_players)
public_state.__turn__ = (public_state.turn + 1) % public_state.param_num_normal_players
else:
logger.fatal("BangEnv generates %s action for responding Indian"%(action.key))
raise Exception("BangEnv generates %s action for responding Indian"%(action.key))
elif isinstance(response_action,BangAction) == True \
and response_action.type == BangActionType.card \
and response_action.card.name == PlayingCardNames.Catling:
if action.type == BangActionType.other and action.other == OtherActionNames.giveup:
person_states[public_state.__turn__].__hp__ -= 1
elif action.type == BangActionType.card and action.card.name == PlayingCardNames.Miss:
person_states.__hand_cards__.remove(action.card)
new_turn = (public_state.turn + 1) % (public_state.param_num_normal_players)
public_state.__turn__ = (public_state.turn + 1) % public_state.param_num_normal_players
else:
logger.fatal("BangEnv generates %s action for responding Indian" % (action.key))
raise Exception("BangEnv generates %s action for responding Indian" % (action.key))
def lookup(cls, key):
logger = roomai.get_logger()
if key not in AllCharacterCardsDict:
logger.fatal("key (%s) is not invalid charactor key" % (key))
raise ValueError("key (%s) is not invalid charactor key" % (key))
return AllCharacterCardsDict[key]
def init(self, params=dict()):
"""
Args:
params:
Returns:
"""
self.logger = roomai.get_logger()
if "num_players" in params:
self.num_players = params["num_players"]
else:
self.num_players = 3
if "dealer_id" in params:
self.dealer_id = params["dealer_id"]
else:
self.dealer_id = int(random.random() * self.num_players)
if "chips" in params:
self.chips = params["chips"]
else:
self.chips = [1000 for i in range(self.num_players)]
if "big_blind_bet" in params:
self.big_blind_bet = params["big_blind_bet"]
else:
self.big_blind_bet = 10
if "allcards" in params:
self.allcards = [c.__deepcopy__() for c in params["allcards"]]
else:
self.allcards = []
for i in xrange(13):
for j in xrange(4):
self.allcards.append(roomai.common.PokerCard(i, j))
random.shuffle(self.allcards)
if "record_history" in params:
self.record_history = params["record_history"]
else:
self.record_history = False
self.check_initialization_configuration(self)
hand_cards = []
for i in xrange(self.num_players):
hand_cards.append(self.allcards[i * 2:(i + 1) * 2])
keep_cards = self.allcards[self.num_players * 2:self.num_players * 2 +
5]
## public info
small = (self.dealer_id + 1) % self.num_players
big = (self.dealer_id + 2) % self.num_players
self.public_state = TexasHoldemPublicState()
self.public_state.num_players = self.num_players
self.public_state.dealer_id = self.dealer_id
self.public_state.big_blind_bet = self.big_blind_bet
self.public_state.raise_account = self.big_blind_bet
self.public_state.is_fold = [False for i in xrange(self.num_players)]
self.public_state.num_quit = 0
self.public_state.is_allin = [False for i in xrange(self.num_players)]
self.public_state.num_allin = 0
self.public_state.is_needed_to_action = [
True for i in xrange(self.num_players)
]
self.public_state.num_needed_to_action = self.public_state.num_players
self.public_state.bets = [0 for i in xrange(self.num_players)]
self.public_state.chips = self.chips
self.public_state.stage = StageSpace.firstStage
self.public_state.turn = (big + 1) % self.public_state.num_players
self.public_state.public_cards = []
self.public_state.previous_id = None
self.public_state.previous_action = None
if self.public_state.chips[big] > self.big_blind_bet:
self.public_state.chips[big] -= self.big_blind_bet
self.public_state.bets[big] += self.big_blind_bet
else:
self.public_state.bets[big] = self.public_state.chips[big]
self.public_state.chips[big] = 0
self.public_state.is_allin[big] = True
self.public_state.num_allin += 1
self.public_state.max_bet_sofar = self.public_state.bets[big]
self.public_state.raise_account = self.big_blind_bet
if self.public_state.chips[small] > self.big_blind_bet / 2:
self.public_state.chips[small] -= self.big_blind_bet / 2
self.public_state.bets[small] += self.big_blind_bet / 2
else:
self.public_state.bets[small] = self.public_state.chips[small]
self.public_state.chips[small] = 0
self.public_state.is_allin[small] = True
self.public_state.num_allin += 1
self.public_state.is_terminal = False
self.public_state.scores = None
# private info
self.private_state = TexasHoldemPrivateState()
self.private_state.hand_cards = [[] for i in xrange(self.num_players)]
for i in xrange(self.num_players):
self.private_state.hand_cards[i] = [
hand_cards[i][j].__deepcopy__()
for j in xrange(len(hand_cards[i]))
]
self.private_state.keep_cards = [
keep_cards[i].__deepcopy__() for i in xrange(len(keep_cards))
]
## person info
self.person_states = [
TexasHoldemPersonState() for i in xrange(self.num_players)
]
for i in xrange(self.num_players):
self.person_states[i].id = i
self.person_states[i].hand_cards = [
hand_cards[i][j].__deepcopy__()
for j in xrange(len(hand_cards[i]))
]
self.person_states[
self.public_state.turn].available_actions = self.available_actions(
self.public_state, self.person_states[self.public_state.turn])
self.__gen_history__()
infos = self.__gen_infos__()
if self.logger.level <= logging.DEBUG:
self.logger.debug("TexasHoldemEnv.init: num_players = %d, dealer_id = %d, chip = %d, big_blind_bet = %d"%(\
self.public_state.num_players,\
self.public_state.dealer_id,\
self.public_state.chips[0],\
self.public_state.big_blind_bet
))
return infos, self.public_state, self.person_states, self.private_state