def test_get_possible_range_idxs_leduc(self):
for n in range(2, 9):
env_bldr = get_leduc_env_bldr()
# if actually blocked
for c in range(env_bldr.rules.N_CARDS_IN_DECK):
board_2d = env_bldr.lut_holder.get_2d_cards(
np.array([c], dtype=np.int32))
result = PokerRange.get_possible_range_idxs(
rules=env_bldr.rules,
lut_holder=env_bldr.lut_holder,
board_2d=board_2d)
should_be = np.delete(
np.arange(env_bldr.rules.RANGE_SIZE, dtype=np.int32), c)
assert np.array_equal(a1=result, a2=should_be)
# if nothing blocked
board_2d = np.array([Poker.CARD_NOT_DEALT_TOKEN_2D], dtype=np.int8)
result = PokerRange.get_possible_range_idxs(
rules=env_bldr.rules,
lut_holder=env_bldr.lut_holder,
board_2d=board_2d)
should_be = np.arange(env_bldr.rules.RANGE_SIZE, dtype=np.int32)
assert np.array_equal(a1=result, a2=should_be)
def test_normalize(self):
env_bldr = get_leduc_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
range_._range = np.random.random(size=env_bldr.rules.RANGE_SIZE)
range_.normalize()
np.testing.assert_allclose(np.sum(range_._range), 1, atol=0.0001)
def test_normalize_all_zero(self):
env_bldr = get_leduc_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
range_._range = np.zeros_like(range_._range)
range_.normalize()
np.testing.assert_allclose(np.sum(range_._range), 1, atol=0.0001)
def test_get_new_blockers_1d_leduc(self):
env_bldr = get_leduc_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
full_board = np.array([[2, 1]], dtype=np.int8)
should_be = {
Poker.PREFLOP: env_bldr.lut_holder.get_1d_cards(full_board[:0]),
Poker.FLOP: env_bldr.lut_holder.get_1d_cards(full_board),
}
for _round in [Poker.FLOP]:
_n = env_bldr.lut_holder.DICT_LUT_N_CARDS_OUT[
Poker.FLOP] - env_bldr.lut_holder.DICT_LUT_N_CARDS_OUT[_round]
if _round == Poker.FLOP:
board_2d = np.copy(full_board)
else:
board_2d = np.concatenate(
(full_board[:env_bldr.lut_holder.
DICT_LUT_N_CARDS_OUT[_round]],
np.array(
[Poker.CARD_NOT_DEALT_TOKEN_2D for _ in range(_n)],
dtype=np.int8)))
result = range_._get_new_blockers_1d(game_round=_round,
board_2d=board_2d)
assert np.array_equal(a1=result, a2=should_be[_round])
def test_get_new_blockers_1d_holdem(self):
env_bldr = get_holdem_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
full_board = np.array([[1, 2], [3, 3], [12, 1], [5, 2], [6, 0]],
dtype=np.int8)
should_be = {
Poker.PREFLOP: env_bldr.lut_holder.get_1d_cards(full_board[:0]),
Poker.FLOP: env_bldr.lut_holder.get_1d_cards(full_board[0:3]),
Poker.TURN: env_bldr.lut_holder.get_1d_cards(full_board[3:4]),
Poker.RIVER: env_bldr.lut_holder.get_1d_cards(full_board[4:5]),
}
for _round in [Poker.PREFLOP, Poker.FLOP, Poker.TURN, Poker.RIVER]:
_n = env_bldr.lut_holder.DICT_LUT_N_CARDS_OUT[Poker.RIVER] - \
env_bldr.lut_holder.DICT_LUT_N_CARDS_OUT[_round]
if _round == Poker.RIVER:
board_2d = np.copy(full_board)
else:
board_2d = np.concatenate(
(full_board[:env_bldr.lut_holder.
DICT_LUT_N_BOARD_BRANCHES[_round]],
np.array(
[Poker.CARD_NOT_DEALT_TOKEN_2D for _ in range(_n)],
dtype=np.int8)))
result = range_._get_new_blockers_1d(game_round=_round,
board_2d=board_2d)
assert np.array_equal(a1=result, a2=should_be[_round])
def test_get_card_probs_holdem(self):
env_bldr = get_holdem_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
cards_to_remove = np.array([0, 3, 6, 33, 21, 51], np.int8)
# use previously tested method to make this test easier
range_.set_cards_to_zero_prob(
cards_2d=env_bldr.lut_holder.get_2d_cards(cards_to_remove))
r = range_.get_card_probs()
assert np.allclose(np.sum(r), 2, atol=0.00001)
for c in cards_to_remove:
assert np.allclose(r[c], 0, atol=0.00001)
def test_remove_cards_from_raw_range_holdem(self):
env_bldr = get_holdem_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
holdem_cards = np.array([[7, 2], [6, 0]], dtype=np.int8)
range_.set_cards_to_zero_prob(cards_2d=holdem_cards)
_ra = range_._range.reshape(-1, env_bldr.rules.RANGE_SIZE)
for i in range(_ra.shape[0]):
np.testing.assert_allclose(np.sum(_ra[i]), 1, atol=0.00001)
_assert_cards_not_in_ranges(cards_2d=holdem_cards,
ranges=_ra,
rules=env_bldr.rules,
lut_holder=env_bldr.lut_holder)
def _fill_chance_node_strategy(self, node):
assert node.strategy is None
if node.is_terminal:
return
if node.p_id_acting_next == self._tree.CHANCE_ID:
game_round = node.children[0].env_state[EnvDictIdxs.current_round]
n_children = len(node.children)
assert n_children == self._env_bldr.lut_holder.DICT_LUT_N_BOARDS[
game_round]
# chance nodes are uniform random
node.strategy = np.zeros(shape=(self._env_bldr.rules.RANGE_SIZE,
n_children),
dtype=np.float32)
# set strategy for impossible hands to 0
for c_id in range(n_children):
mask = PokerRange.get_possible_range_idxs(
rules=self._env_bldr.rules,
lut_holder=self._env_bldr.lut_holder,
board_2d=node.children[c_id].env_state[
EnvDictIdxs.board_2d])
node.strategy[
mask,
c_id] = 1.0 / (self._env_bldr.rules.N_CARDS_IN_DECK - 2)
for c in node.children:
self._fill_chance_node_strategy(node=c)
def __init__(self, t_prof, chief_handle, eval_agent_cls):
assert t_prof.n_seats == 2
self.t_prof = t_prof
self.lbr_args = t_prof.module_args["lbr"]
self._eval_env_bldr = _util.get_env_builder_lbr(t_prof=t_prof)
self.check_to_round = self.lbr_args.lbr_check_to_round
self.chief_handle = chief_handle
self.agent = _AgentWrapper(t_prof=t_prof, lbr_args=self.lbr_args, eval_agent_cls=eval_agent_cls)
# has different raise sizes than agent's env! This needs to be considered when updating the envs after opp acts
self._env = None
self.agent_range = PokerRange(env_bldr=self._eval_env_bldr)
assert self.check_to_round is None or (self.check_to_round in self._eval_env_bldr.rules.ALL_ROUNDS_LIST)
def test_create(self):
env_bldr = get_leduc_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
assert np.allclose(np.sum(range_._range.reshape(
-1, env_bldr.rules.RANGE_SIZE),
axis=1),
1,
atol=0.0001)
def test_get_possible_range_idxs_holdem(self):
env_bldr = get_holdem_env_bldr()
for n in range(2, 9):
board_2d = np.array(
[[0, 0], [5, 2], [12, 3], Poker.CARD_NOT_DEALT_TOKEN_2D],
dtype=np.int8)
result = PokerRange.get_possible_range_idxs(
rules=env_bldr.rules,
lut_holder=env_bldr.lut_holder,
board_2d=board_2d)
assert result.shape[0] == 1176 # nck of 49:2
# all of these should be blocked
for e in [0, 1, 2, 3, 4, 50, 1325]:
assert not np.any(result == e)
def get_n_board_branches_LUT(self):
_N_CARDS_DEALT_IN_TRANSITION_TO_LUT = self.get_n_cards_dealt_in_transition_to_LUT(
)
_N_CARDS_OUT_AT = self.get_n_cards_out_at_LUT()
lut = {Poker.PREFLOP: 0}
for r in [
_r for _r in self.rules.ALL_ROUNDS_LIST if _r != Poker.PREFLOP
]:
nc = self.rules.N_CARDS_IN_DECK \
- _N_CARDS_OUT_AT[self.rules.ROUND_BEFORE[r]] \
- self.rules.N_HOLE_CARDS
# get_range_size is actually a general combinatorial function that we can also use here
lut[r] = PokerRange.get_range_size(
n_hole_cards=_N_CARDS_DEALT_IN_TRANSITION_TO_LUT[r],
n_cards_in_deck=nc)
return lut
def test_get_range_size(self):
assert PokerRange.get_range_size(n_hole_cards=2,
n_cards_in_deck=52) == 1326
assert PokerRange.get_range_size(n_hole_cards=4,
n_cards_in_deck=52) == 270725
assert PokerRange.get_range_size(n_hole_cards=1,
n_cards_in_deck=52) == 52
assert PokerRange.get_range_size(n_hole_cards=1,
n_cards_in_deck=6) == 6
assert PokerRange.get_range_size(n_hole_cards=1,
n_cards_in_deck=6) == 6
assert PokerRange.get_range_size(n_hole_cards=3,
n_cards_in_deck=6) == 20
class LocalLBRWorker:
"""
Slave to EvalLBRMaster. Does the LBR computation as described in https://arxiv.org/abs/1612.07547
"""
def __init__(self, t_prof, chief_handle, eval_agent_cls):
assert t_prof.n_seats == 2
self.t_prof = t_prof
self.lbr_args = t_prof.module_args["lbr"]
self._eval_env_bldr = _util.get_env_builder_lbr(t_prof=t_prof)
self.check_to_round = self.lbr_args.lbr_check_to_round
self.chief_handle = chief_handle
self.agent = _AgentWrapper(t_prof=t_prof,
lbr_args=self.lbr_args,
eval_agent_cls=eval_agent_cls)
# has different raise sizes than agent's env! This needs to be considered when updating the envs after opp acts
self._env = None
self.agent_range = PokerRange(env_bldr=self._eval_env_bldr)
assert self.check_to_round is None or (
self.check_to_round in self._eval_env_bldr.rules.ALL_ROUNDS_LIST)
def run(self, agent_seat_id, n_iterations, mode, stack_size):
""" returns an estimate of a lower bound of the exploitablity of the agent """
self.agent.set_mode(mode=mode)
self.agent.to_stack_size(stack_size)
self.agent_range.reset()
self._env = self._eval_env_bldr.get_new_env(is_evaluating=True,
stack_size=stack_size)
if not self.agent.can_compute_mode():
return None
if self._eval_env_bldr.env_cls.IS_FIXED_LIMIT_GAME:
return self._run_limit(agent_seat_id=agent_seat_id,
n_iterations=n_iterations)
else:
return self._run_no_limit(agent_seat_id=agent_seat_id,
n_iterations=n_iterations)
def update_weights(self, weights_for_eval_agent):
self.agent.update_weights(weights_for_eval_agent)
def _reset_episode(self):
ret = self._env.reset()
self.agent.reset(deck_state_dict=self._env.cards_state_dict())
self.agent_range.reset()
return ret
def _run_limit(self, agent_seat_id, n_iterations):
total_lbr_winnings = np.empty(shape=n_iterations, dtype=np.float32)
lbr_seat_id = 1 - agent_seat_id
for iteration_id in range(n_iterations):
if iteration_id % 50 == 0:
print("LBR hand: ", iteration_id)
# """""""""""""""""
# Reset
# """""""""""""""""
env_obs, reward, terminal, info = self._reset_episode()
lbr_hand = self._env.get_hole_cards_of_player(p_id=lbr_seat_id)
self.agent_range.set_cards_to_zero_prob(cards_2d=lbr_hand)
# """""""""""""""""
# Play Episode
# """""""""""""""""
while not terminal:
p_id_acting = self._env.current_player.seat_id
if self.t_prof.DEBUGGING:
assert p_id_acting == self.agent.cpu_agent._internal_env_wrapper.env.current_player.seat_id
if p_id_acting == lbr_seat_id:
# optional feature: check the first N rounds 100% as LBR
if (self.check_to_round is not None) and (
self._env.current_round < self.check_to_round):
action_int = Poker.CHECK_CALL
else:
_rollout_mngr = _LBRRolloutManager(
t_prof=self.t_prof,
env_bldr=self._eval_env_bldr,
env=self._env,
lbr_hand_2d=lbr_hand)
# illegal: -1, fold: 0, all other: any float
_utility = np.full(shape=3,
fill_value=-1.0,
dtype=np.float32)
# ev(s, lbr_a=fold)
_utility[Poker.FOLD] = 0.0
# ev(s, lbr_a=check_call)
_wp = _rollout_mngr.get_lbr_checkdown_equity(
agent_range=self.agent_range
) # if check/called down
_asked = self._env.seats[
agent_seat_id].current_bet - self._env.seats[
lbr_seat_id].current_bet
_pot_before_action = self._env.get_all_winnable_money()
_utility[
Poker.CHECK_CALL] = _wp * _pot_before_action - (
1 - _wp) * _asked
# prepare for raise simulation
if Poker.BET_RAISE in self._env.get_legal_actions():
_saved_env_state = self._env.state_dict()
_saved_agent_env_state = self.agent.env_state_dict(
)
_saved_agent_range_state = self.agent_range.state_dict(
)
# compute ev for raise
# _________________________________ simulate LBR play r ____________________________________
self._env.step(action=Poker.BET_RAISE)
_pot_after_raise = self._env.get_all_winnable_money(
)
self.agent.notify_of_action(
p_id_acted=lbr_seat_id,
action_he_did=Poker.BET_RAISE)
# what agent would do after LBR raises. DOESN'T STEP INTERNAL ENV!
_, a_probs_each_hand = self.agent.get_action(
step_env=False, need_probs=True)
# _______________________________ simulate agent reaction __________________________________
# p(agent_fold)
_fold_prob = np.sum(
self.agent_range.range *
a_probs_each_hand[:, Poker.FOLD])
# p(not agent_fold | hand)
_p_not_fold_per_hand = (
1 - a_probs_each_hand[:, Poker.FOLD])
# agent_range after not folding
self.agent_range.mul_and_norm(_p_not_fold_per_hand)
# p(lbr_win | lbr play r -> agent play not fold)
_wp_now = _rollout_mngr.get_lbr_checkdown_equity(
agent_range=self.agent_range)
# ev(state, lbr_a=r)
_chips_lbr_puts_in_pot = _pot_after_raise - _pot_before_action
_ev_if_fold = _pot_before_action
_ev_if_not_fold = (_wp_now * _pot_after_raise) - (
(1 - _wp_now) * _chips_lbr_puts_in_pot)
_utility[
Poker.BET_RAISE] = _fold_prob * _ev_if_fold + (
1 - _fold_prob) * _ev_if_not_fold
# ________________________________________ reset ___________________________________________
self.agent_range.load_state_dict(
_saved_agent_range_state)
self._env.load_state_dict(_saved_env_state)
self.agent.load_env_state_dict(
_saved_agent_env_state)
# select action with highest approximated EV
action_int = np.argmax(_utility)
# ________________________________________ notify agent ____________________________________________
self.agent.notify_of_action(p_id_acted=lbr_seat_id,
action_he_did=action_int)
else: # agent has to act
action_int, a_probs_each_hand = self.agent.get_action(
step_env=True, need_probs=True)
self.agent_range.update_after_action(
action=action_int,
all_a_probs_for_all_hands=a_probs_each_hand)
# _____________________________________________ step ___________________________________________________
old_game_round = self._env.current_round
env_obs, reward, terminal, info = self._env.step(
action=action_int)
if self._env.current_round != old_game_round:
self.agent_range.update_after_new_round(
new_round=self._env.current_round,
board_now_2d=self._env.board)
total_lbr_winnings[iteration_id] = reward[
lbr_seat_id] * self._env.REWARD_SCALAR * self._env.EV_NORMALIZER
return total_lbr_winnings
def _run_no_limit(self, agent_seat_id, n_iterations):
total_lbr_winnings = np.empty(shape=n_iterations, dtype=np.float32)
lbr_seat_id = 1 - agent_seat_id
n_lbr_bets = len(self._env.bet_sizes_list_as_frac_of_pot)
for iteration_id in range(n_iterations):
if iteration_id % 50 == 0:
print("LBR hand: ", iteration_id)
# """""""""""""""""
# Reset
# """""""""""""""""
env_obs, reward, done, info = self._reset_episode()
lbr_hand = self._env.get_hole_cards_of_player(p_id=lbr_seat_id)
self.agent_range.set_cards_to_zero_prob(cards_2d=lbr_hand)
# """""""""""""""""
# Play Episode
# """""""""""""""""
while not done:
p_id_acting = self._env.current_player.seat_id
if self.t_prof.DEBUGGING:
assert p_id_acting == self.agent.cpu_agent._internal_env_wrapper.env.current_player.seat_id
if p_id_acting == lbr_seat_id:
# optional feature: check the first N rounds 100% as LBR
if (self.check_to_round is not None) and (
self._env.current_round < self.check_to_round):
action_int = Poker.CHECK_CALL
else:
_rollout_mngr = _LBRRolloutManager(
t_prof=self.t_prof,
env_bldr=self._eval_env_bldr,
env=self._env,
lbr_hand_2d=lbr_hand)
# illegal: -1, fold: 0, all other: any float
_utility = np.full(shape=2 + n_lbr_bets,
fill_value=-1.0,
dtype=np.float32)
# ev(s, lbr_a=fold)
_utility[Poker.FOLD] = 0.0
# ev(s, lbr_a=check_call)
_wp = _rollout_mngr.get_lbr_checkdown_equity(
agent_range=self.agent_range)
_asked = self._env.seats[
agent_seat_id].current_bet - self._env.seats[
lbr_seat_id].current_bet
_pot_before_action = self._env.get_all_winnable_money()
_utility[
Poker.CHECK_CALL] = _wp * _pot_before_action - (
1 - _wp) * _asked
# prepare for raise simulation
_saved_env_state = self._env.state_dict()
_saved_agent_env_state = self.agent.env_state_dict()
_saved_agent_range_state = self.agent_range.state_dict(
)
_legal_raises = self._env.get_legal_actions()
for a in [Poker.FOLD, Poker.CHECK_CALL]:
if a in _legal_raises:
_legal_raises.remove(a)
# compute ev for all raise sizes LBR can choose from
for r in _legal_raises:
raise_frac = self._env.bet_sizes_list_as_frac_of_pot[
r - 2]
# _________________________________ simulate LBR play r ____________________________________
self._env.step(action=r)
_pot_after_raise = self._env.get_all_winnable_money(
)
self.agent.notify_of_raise_frac_action(
p_id_acted=lbr_seat_id, frac=raise_frac)
if self.t_prof.DEBUGGING:
assert agent_seat_id == self.agent.cpu_agent._internal_env_wrapper.env.current_player.seat_id
# what agent would do after LBR raises. DOESN'T STEP INTERNAL ENV!
a_probs_each_hand = self.agent.get_a_probs_for_each_hand(
)
# _______________________________ simulate agent reaction __________________________________
# p(agent_fold)
_fold_prob = np.sum(
self.agent_range.range *
a_probs_each_hand[:, Poker.FOLD])
# p(not agent_fold | hand)
_p_not_fold_per_hand = (
1 - a_probs_each_hand[:, Poker.FOLD])
# agent_range after not folding
self.agent_range.mul_and_norm(_p_not_fold_per_hand)
# p(lbr_win | lbr play r -> agent play not fold)
_wp_now = _rollout_mngr.get_lbr_checkdown_equity(
agent_range=self.agent_range)
# ev(state, lbr_a=r)
_chips_lbr_puts_in_pot = _pot_after_raise - _pot_before_action
_ev_if_fold = _pot_before_action
_ev_if_not_fold = (_wp_now * _pot_after_raise) - (
(1 - _wp_now) * _chips_lbr_puts_in_pot)
_utility[r] = _fold_prob * _ev_if_fold + (
1 - _fold_prob) * _ev_if_not_fold
# ________________________________________ reset ___________________________________________
self.agent_range.load_state_dict(
_saved_agent_range_state)
self._env.load_state_dict(_saved_env_state)
self.agent.load_env_state_dict(
_saved_agent_env_state)
# select action with highest approximated EV
action_int = np.argmax(_utility)
# ________________________________________ notify agent ____________________________________________
if action_int >= 2:
raise_frac = self._env.bet_sizes_list_as_frac_of_pot[
action_int - 2]
self.agent.notify_of_raise_frac_action(
p_id_acted=lbr_seat_id, frac=raise_frac)
else:
self.agent.notify_of_action(p_id_acted=lbr_seat_id,
action_he_did=action_int)
else: # agent has to act
if self.t_prof.DEBUGGING:
assert p_id_acting == self.agent.cpu_agent._internal_env_wrapper.env.current_player.seat_id
action_int, a_probs_each_hand = self.agent.get_action(
step_env=True, need_probs=True)
self.agent_range.update_after_action(
action=action_int,
all_a_probs_for_all_hands=a_probs_each_hand)
if action_int >= 2:
# querying what the bet size is in the agent's env_args (these might differ from LBR's!).
raise_frac = \
self.agent.cpu_agent.env_bldr.env_args.bet_sizes_list_as_frac_of_pot[action_int - 2]
# _____________________________________________ step ___________________________________________________
old_game_round = self._env.current_round
if action_int >= 2: # step with fraction because agent and LBR have different raise sizes
env_obs, reward, done, info = self._env.step_raise_pot_frac(
pot_frac=raise_frac)
else:
env_obs, reward, done, info = self._env.step(
action=action_int)
if self._env.current_round != old_game_round:
self.agent_range.update_after_new_round(
new_round=self._env.current_round,
board_now_2d=self._env.board)
total_lbr_winnings[iteration_id] = reward[
lbr_seat_id] * self._env.REWARD_SCALAR * self._env.EV_NORMALIZER
return total_lbr_winnings
class Flop3HoldemRules:
"""
General rules of Texas Hold'em poker games
"""
N_HOLE_CARDS = 2
N_RANKS = 13
N_SUITS = 4
N_CARDS_IN_DECK = N_RANKS * N_SUITS
RANGE_SIZE = PokerRange.get_range_size(n_hole_cards=N_HOLE_CARDS, n_cards_in_deck=N_CARDS_IN_DECK)
BTN_IS_FIRST_POSTFLOP = False
N_FLOP_CARDS = 3
N_TURN_CARDS = 0
N_RIVER_CARDS = 0
N_TOTAL_BOARD_CARDS = N_FLOP_CARDS
ALL_ROUNDS_LIST = [Poker.PREFLOP, Poker.FLOP]
SUITS_MATTER = True
ROUND_BEFORE = {
Poker.PREFLOP: Poker.PREFLOP,
Poker.FLOP: Poker.PREFLOP,
Poker.TURN: None,
Poker.RIVER: None,
}
ROUND_AFTER = {
Poker.PREFLOP: Poker.FLOP,
Poker.FLOP: None,
Poker.TURN: None,
Poker.RIVER: None,
}
RANK_DICT = {
Poker.CARD_NOT_DEALT_TOKEN_1D: "",
0: "2",
1: "3",
2: "4",
3: "5",
4: "6",
5: "7",
6: "8",
7: "9",
8: "T",
9: "J",
10: "Q",
11: "K",
12: "A"
}
SUIT_DICT = {
Poker.CARD_NOT_DEALT_TOKEN_1D: "",
0: "h",
1: "d",
2: "s",
3: "c"
}
STRING = "FLOP3_HOLDEM_RULES"
def __init__(self):
from PokerRL.game._.cpp_wrappers.CppHandeval import CppHandeval
self._clib = CppHandeval()
def get_hand_rank_all_hands_on_given_boards(self, boards_1d, lut_holder):
"""
for docs refer to PokerEnv
"""
raise NotImplementedError(
"Batched hand eval for multiple boards is unfortunately not implemented for FHP at the moment."
) # TODO
def get_hand_rank(self, hand_2d, board_2d):
"""
for docs refer to PokerEnv
"""
return self._clib.get_hand_rank_fhp3(hand_2d=hand_2d, board_2d=board_2d)
@classmethod
def get_lut_holder(cls):
from PokerRL.game._.look_up_table import LutHolderHoldem
return LutHolderHoldem(cls)
class BigLeducRules:
N_HOLE_CARDS = 1
N_RANKS = 12
N_SUITS = 2
N_CARDS_IN_DECK = N_RANKS * N_SUITS
RANGE_SIZE = PokerRange.get_range_size(n_hole_cards=N_HOLE_CARDS, n_cards_in_deck=N_CARDS_IN_DECK)
BTN_IS_FIRST_POSTFLOP = True
N_FLOP_CARDS = 1
N_TURN_CARDS = 0
N_RIVER_CARDS = 0
N_TOTAL_BOARD_CARDS = N_FLOP_CARDS + N_TURN_CARDS + N_RIVER_CARDS
ALL_ROUNDS_LIST = [Poker.PREFLOP, Poker.FLOP]
SUITS_MATTER = False
ROUND_BEFORE = {
Poker.PREFLOP: Poker.PREFLOP,
Poker.FLOP: Poker.PREFLOP
}
ROUND_AFTER = {
Poker.PREFLOP: Poker.FLOP,
Poker.FLOP: None
}
RANK_DICT = {i: str(i + 2) for i in range(N_RANKS)}
SUIT_DICT = {k: ["a", "b", "c", "d", "e", "f", "g"][k] for k in range(N_SUITS)} \
if N_SUITS < 8 \
else {i: "_" + str(i) for i in range(N_SUITS)}
STRING = "BIG_LEDUC_RULES"
def __init__(self):
pass
def get_hand_rank_all_hands_on_given_boards(self, boards_1d, lut_holder):
"""
for general docs refer to PokerEnv
"""
hand_ranks = np.full(shape=(boards_1d.shape[0], LeducRules.RANGE_SIZE), fill_value=-1, dtype=np.int32)
for board_idx in range(boards_1d.shape[0]):
for range_idx in range(LeducRules.RANGE_SIZE):
hand_ranks[board_idx, range_idx] = self.get_hand_rank(
hand_2d=lut_holder.get_2d_hole_cards_from_range_idx(range_idx=range_idx),
board_2d=lut_holder.get_2d_cards(cards_1d=boards_1d[board_idx]))
return hand_ranks
def get_hand_rank(self, hand_2d, board_2d):
"""
for docs refer to PokerEnv
"""
if board_2d[0, 0] == hand_2d[0, 0]:
return 10000 + hand_2d[0, 0]
else:
return hand_2d[0, 0]
@classmethod
def get_lut_holder(cls):
from PokerRL.game._.look_up_table import LutHolderLeduc
return LutHolderLeduc(cls)
def test_save_load(self):
env_bldr = get_leduc_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
range_.load_state_dict(range_.state_dict())
def test_get_card_probs_leduc(self):
env_bldr = get_leduc_env_bldr()
range_ = PokerRange(env_bldr=env_bldr)
assert np.array_equal(range_.get_card_probs(), range_._range)
class PLORules:
"""
General rules of Pot Limit Omaha
"""
N_HOLE_CARDS = 4
N_RANKS = 13
N_SUITS = 4
N_CARDS_IN_DECK = N_RANKS * N_SUITS
RANGE_SIZE = PokerRange.get_range_size(n_hole_cards=N_HOLE_CARDS, n_cards_in_deck=N_CARDS_IN_DECK)
BTN_IS_FIRST_POSTFLOP = False
N_FLOP_CARDS = 3
N_TURN_CARDS = 1
N_RIVER_CARDS = 1
N_TOTAL_BOARD_CARDS = N_FLOP_CARDS + N_TURN_CARDS + N_RIVER_CARDS
ALL_ROUNDS_LIST = [Poker.PREFLOP, Poker.FLOP, Poker.TURN, Poker.RIVER]
SUITS_MATTER = True
ROUND_BEFORE = {
Poker.PREFLOP: Poker.PREFLOP,
Poker.FLOP: Poker.PREFLOP,
Poker.TURN: Poker.FLOP,
Poker.RIVER: Poker.TURN
}
ROUND_AFTER = {
Poker.PREFLOP: Poker.FLOP,
Poker.FLOP: Poker.TURN,
Poker.TURN: Poker.RIVER,
Poker.RIVER: None
}
RANK_DICT = {
Poker.CARD_NOT_DEALT_TOKEN_1D: "",
0: "2",
1: "3",
2: "4",
3: "5",
4: "6",
5: "7",
6: "8",
7: "9",
8: "T",
9: "J",
10: "Q",
11: "K",
12: "A"
}
SUIT_DICT = {
Poker.CARD_NOT_DEALT_TOKEN_1D: "",
0: "h",
1: "d",
2: "s",
3: "c"
}
STRING = "PLO_RULES"
def __init__(self):
from PokerRL.game._.cpp_wrappers.CppHandeval import CppHandeval
self._clib = CppHandeval()
def get_hand_rank_all_hands_on_given_boards(self, boards_1d, lut_holder):
"""
for docs refer to PokerEnv
returns a numpy array [1,hole_hands_total] where best hand has biggest number
and not possible hand has -1
"""
r = self._clib.get_hand_rank_all_hands_on_given_boards_52_holdem(boards_1d=boards_1d, lut_holder=lut_holder)
return r
def get_hand_rank(self, hand_2d, board_2d):
"""
for docs refer to PokerEnv
"""
r = self._clib.get_hand_rank_52_plo(hand_2d=hand_2d, board_2d=board_2d)
return r
@classmethod
def get_lut_holder(cls):
from PokerRL.game._.look_up_table import LutHolderPLO
return LutHolderPLO(cls)