def test_three_players(self):
deck = pydealer.Deck()
cards = deck.get_list([
'10 of Spades', 'Jack of Spades', 'Queen of Spades',
'King of Spades'
])
t, j, q, k = cards
players = create_player_set(3)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
TKQ_node = root.get_children()[(t, k, q)]
JKQ_node = root.get_children()[(j, k, q)]
for node in [TKQ_node, JKQ_node]:
self.assertTrue(
np.allclose(
node.play(PokerActions.RAISE_1).play(
PokerActions.FOLD).play(
PokerActions.FOLD).evaluation(),
np.array([2, -1, -1])))
self.assertTrue(
np.allclose(
node.play(PokerActions.CHECK).play(
PokerActions.RAISE_1).play(PokerActions.FOLD).play(
PokerActions.CALL).evaluation(),
np.array([-2, 3, -1])))
def run_game():
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
#random.seed(1)
#chance_sampling_cfr = ChanceSamplingCFR(root, players)
#chance_sampling_cfr.run(iterations=1000)
#chance_sampling_cfr.compute_nash_equilibrium()
# read Nash-Equilibrum via chance_sampling_cfr.nash_equilibrium member
# try chance_sampling_cfr.value_of_the_game() function to get value of the game (-1/18)
#print(chance_sampling_cfr.value_of_the_game())
# vanilla cfr
vanilla_cfr = VanillaCFR(root, players)
root = game.create_root_node()
j, q, k = cards
KQ_node = root.get_children()[(k, q)]
vanilla_cfr._cfr_utility_recursive(
KQ_node.play(PokerActions.CHECK).play(PokerActions.CHECK), [1, 1])
for i in tqdm(range(10)):
vanilla_cfr.run(iterations=100)
vanilla_cfr.compute_nash_equilibrium()
print(vanilla_cfr.value_of_the_game())
def run_game():
deck = pydealer.Deck()
hand = pydealer.Stack()
cards = deck.get_list([
'10 of Spades', 'Jack of Spades', 'Queen of Spades', 'King of Spades'
])
players = create_player_set(3)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
#random.seed(1)
#chance_sampling_cfr = ChanceSamplingCFR(root, players)
#chance_sampling_cfr.run(iterations=1000)
#chance_sampling_cfr.compute_nash_equilibrium()
# read Nash-Equilibrum via chance_sampling_cfr.nash_equilibrium member
# try chance_sampling_cfr.value_of_the_game() function to get value of the game (-1/18)
#print(chance_sampling_cfr.value_of_the_game())
# vanilla cfr
vanilla_cfr = VanillaCFR(root, players)
for i in tqdm(range(10)):
vanilla_cfr.run(iterations=100)
vanilla_cfr.compute_nash_equilibrium()
def get_chance_node(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
return game.create_root_node()
def test_evaluation(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
j, q, k = cards
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
KQ_node = root.get_children()[(k, q)]
QJ_node = root.get_children()[(q, j)]
KJ_node = root.get_children()[(k, j)]
QK_node = root.get_children()[(q, k)]
JQ_node = root.get_children()[(j, q)]
JK_node = root.get_children()[(j, k)]
for node in [KQ_node, QJ_node, KJ_node]:
self.assertTrue(
np.allclose(
node.play(PokerActions.RAISE_1).play(
PokerActions.FOLD).evaluation(), np.array([1, -1])))
self.assertTrue(
np.allclose(
node.play(PokerActions.RAISE_1).play(
PokerActions.CALL).evaluation(), np.array([2, -2])))
self.assertTrue(
np.allclose(
node.play(PokerActions.CHECK).play(
PokerActions.RAISE_1).play(
PokerActions.FOLD).evaluation(), np.array([-1,
1])))
self.assertTrue(
np.allclose(
node.play(PokerActions.CHECK).play(
PokerActions.CHECK).evaluation(), np.array([1, -1])))
for node in [QK_node, JQ_node, JK_node]:
self.assertTrue(
np.allclose(
node.play(PokerActions.RAISE_1).play(
PokerActions.FOLD).evaluation(), np.array([1, -1])))
self.assertTrue(
np.allclose(
node.play(PokerActions.RAISE_1).play(
PokerActions.CALL).evaluation(), np.array([-2, 2])))
self.assertTrue(
np.allclose(
node.play(PokerActions.CHECK).play(
PokerActions.RAISE_1).play(
PokerActions.FOLD).evaluation(), np.array([-1,
1])))
self.assertTrue(
np.allclose(
node.play(PokerActions.CHECK).play(
PokerActions.CHECK).evaluation(), np.array([-1, 1])))
def test_player_a_acts_first(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
for k in root.get_children():
child = root.get_children()[k]
assert child.get_player_to_move() == players[0]
def fresh_kq_node(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
vanilla_cfr = VanillaCFR(root, players)
j, q, k = cards
KQ_node = root.get_children()[(k, q)]
return vanilla_cfr, KQ_node
def test_game_tree(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
num_players = 2
game = KuhnGame(create_player_set(num_players), cards, 1)
# see if simulations run successfully
for i in range(100):
curr_node = game.create_root_node()
while not curr_node.is_terminal():
curr_node = random.choice(
list(curr_node.get_children().values()))
self.assertTrue(
len(np.nonzero(curr_node.evaluation())[0]) == num_players)
def fresh_tkq_node():
deck = pydealer.Deck()
cards = deck.get_list([
'10 of Spades', 'Jack of Spades', 'Queen of Spades', 'King of Spades'
])
players = create_player_set(3)
game = KuhnGame(players, cards, num_deal=1)
root = game.create_root_node()
vanilla_cfr = VanillaCFR(root, players)
t, j, q, k = cards
TKQ_node = root.get_children()[(t, k, q)]
return vanilla_cfr, TKQ_node
def test_external_sampling(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
players = create_player_set(2)
game = KuhnGame(players, cards, num_deal=1)
root = game.create_root_node()
chance_cfr = ExternalSamplingCFR(root, players)
chance_cfr.run(iterations=1000)
game_value = chance_cfr.approximate_value_of_game(1000)
epsilon = 0.01
self.assertTrue(game_value[0] < -1. / 18 + epsilon)
self.assertTrue(game_value[1] > 1. / 18 - epsilon)
def test_external_sampling_2(self):
deck = pydealer.Deck()
cards = deck.get_list([
'10 of Spades', 'Jack of Spades', 'Queen of Spades',
'King of Spades'
])
players = create_player_set(3)
game = KuhnGame(players, cards, num_deal=1)
root = game.create_root_node()
chance_cfr = ExternalSamplingCFR(root, players)
chance_cfr.run(iterations=2000)
chance_cfr.compute_nash_equilibrium()
game_value = chance_cfr.approximate_value_of_game(1000)
self.assertTrue(game_value[0] > -3. / 48)
self.assertTrue(game_value[0] < -1. / 48)
self.assertTrue(4. / 48 > game_value[2] > 2. / 48)
def run_game():
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
#players = create_player_set(2)
#ldgame = LDGame(players, 1)
#root = ldgame.create_root_node()
cfr = ExternalSamplingCFR(root, players)
for i in tqdm(range(1000)):
cfr.run(iterations=1)
#if i % 100 == 0:
# print(vanilla_cfr.compute_nash_equilibrium())
print("Approx Value", cfr.approximate_value_of_game(1000))
print(cfr._learned_strategy)
def test_full_iteration(self):
deck = pydealer.Deck()
cards = deck.get_list([
'10 of Spades', 'Jack of Spades', 'Queen of Spades',
'King of Spades'
])
players = create_player_set(3)
game = KuhnGame(players, cards, num_deal=1)
root = game.create_root_node()
vanilla_cfr = VanillaCFR(root, players)
val = np.zeros(3)
for permutation in itertools.permutations(cards, 3):
one_val = vanilla_cfr._cfr_utility_recursive(
root.get_children()[permutation], np.ones(3))
val += one_val
self.assertTrue(
np.allclose(val, np.array([-0.61, 0.26, 0.35]), atol=1E-1))
def test_inf_sets(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
j, q, k = cards
players = create_player_set(2)
game = KuhnGame(players, cards, 1)
root = game.create_root_node()
KQ_node = root.get_children()[(k, q)]
QJ_node = root.get_children()[(q, j)]
KJ_node = root.get_children()[(k, j)]
QK_node = root.get_children()[(q, k)]
JQ_node = root.get_children()[(j, q)]
JK_node = root.get_children()[(j, k)]
assert root.inf_set() == "."
self.assertTrue('King of Spades' in KQ_node.inf_set())
self.assertTrue(
'Queen of Spades' in KQ_node.play(PokerActions.RAISE_1).inf_set())
def test_tjk_12(self):
deck = pydealer.Deck()
cards = deck.get_list([
'10 of Spades', 'Jack of Spades', 'Queen of Spades',
'King of Spades'
])
players = create_player_set(3)
game = KuhnGame(players, cards, num_deal=1)
root = game.create_root_node()
vanilla_cfr = VanillaCFR(root, players)
t, j, q, k = cards
TJQ_node = root.get_children()[(t, j, q)]
TJK_node = root.get_children()[(t, j, k)]
starting_reach = np.ones(3)
vanilla_cfr._cfr_utility_recursive(TJQ_node, starting_reach)
val = vanilla_cfr._cfr_utility_recursive(
TJK_node.play(PokerActions.RAISE_1).play(PokerActions.FOLD),
starting_reach)
def test_possible_hands(self):
deck = pydealer.Deck()
cards = deck.get_list(
['Jack of Spades', 'Queen of Spades', 'King of Spades'])
game = KuhnGame(create_player_set(2), cards, 1)
self.assertEqual(len(list(game.enumerate_possible_hands())), 6)