def test_game():
player = [10, 10]
dealer = [3]
bet = 50
deck = BjDeck(6)
policy_params = {'dealer_card': dealer[0], 'policy': WEB_POLICY}
a = play_hand(player, dealer[0], deck, bet, fixed_policy, **policy_params)
def dealer_stats_for_given_hand(games: int = 1000000, decks: int = 6) -> tuple:
dealer_stats = {}
for i in range(1, 11):
dealer_stats[str(i)] = {
'17': 0,
'18': 0,
'19': 0,
'20': 0,
'21': 0,
'F': 0
}
deck = BjDeck(decks)
for _ in range(games):
deck.start_game()
start_hand = deck.deal()
dealer = [start_hand]
d_cards = dealer_hand(dealer, deck)
count = count_hand(d_cards)
if count > 21:
count = 'F'
dealer_stats[str(start_hand)][str(
count)] = dealer_stats[str(start_hand)][str(count)] + 1.
raw_count = pd.DataFrame(dealer_stats)
raw_count.index.name = 'end count'
raw_count.columns.name = 'start hand'
total = raw_count / raw_count.sum(axis=0)
return raw_count, total
def player_win_rates_for_start_hands(games: int = 1000000,
decks: int = 6) -> tuple:
player_stats = {}
deck = BjDeck(decks)
policy_params = {'policy': WEB_POLICY}
for _ in range(games):
deck.start_game()
hand = [deck.deal()]
dealer = [deck.deal()]
dealer_index = str(dealer[0])
hand.append(deck.deal())
hand_index = convert_hand_to_index(hand)
policy_params['dealer_card'] = dealer[0]
curr_hand = play_hand(hand, dealer[0], deck, 1, fixed_policy,
**policy_params)
d_cards = dealer_hand(dealer, deck)
results = rewards(curr_hand, d_cards)
for result in results:
if result > 0:
end_game = 'win'
elif result < 0:
end_game = 'lost'
else:
end_game = 'equal'
index = f'{hand_index}_{dealer_index}'
if player_stats.get(index):
player_stats[index][
end_game] = player_stats[index][end_game] + 1
player_stats[index]['total'] = player_stats[index]['total'] + 1
else:
player_stats[index] = {
'win': 0,
'lost': 0,
'equal': 0,
'total': 1,
'player_hand': hand_index,
'dealer': dealer_index
}
player_stats[index][end_game] = 1
raw_count = pd.DataFrame(player_stats)
raw_count.index.name = 'Results'
raw_count.columns.name = 'Player hand_Dealer hand'
total = raw_count.apply(get_percents)
return raw_count, total
def player_stats(games: int = 1000000, decks: int = 6) -> tuple:
p_stats = {}
deck = BjDeck(decks)
for _ in range(games):
deck.start_game()
hand = [deck.deal()]
dealer = [deck.deal()]
hand.append(deck.deal())
hand_index = convert_hand_to_index(hand)
policy_params = {'dealer_card': dealer[0], 'policy': WEB_POLICY}
curr_hand = play_hand(hand, dealer[0], deck, 1, fixed_policy,
**policy_params)
for hand in curr_hand:
fill_stats(hand, hand_index, p_stats)
raw_count = pd.DataFrame(p_stats)
raw_count.index.name = 'end count'
raw_count.columns.name = 'start hand'
total = raw_count / raw_count.sum(axis=0)
return raw_count, total
def main():
b = BjDeck(6)
start_money = 0
end_money = start_money
games = 10000
policy_params = {'policy': WEB_POLICY}
for i in range(games):
print(i)
b.start_game()
final_hand, d_hand, return_money = play_game(
b, [100, 100, 100, 100, 100, 100], fixed_policy, **policy_params)
if final_hand is None or return_money is None:
print("bad input")
if len(final_hand) < 3 or len(return_money) < 3:
print("bad input sum")
for hand in final_hand:
print(hand.cards)
print("dealer", d_hand)
print(return_money)
end_money = end_money + sum(return_money)
print("money", end_money)
print(end_money)
print("average win/lost per game: ", (end_money - start_money) / games)
def test_states():
markov = MDP(HAND_OPTIONS)
policy_params = {'actions': markov.actions,
'actions_prob': markov.actions_probs,
'threshold': 90}
deck = BjDeck(6)
start_money = 0
end_money = start_money
for games in range(10000):
deck.start_game()
final_hand, d_hand, rewards = play_game(deck, [1], random_policy, **policy_params)
end_money = end_money + rewards[0]
markov.add_game(final_hand[0], rewards[0])
curr_win = (end_money - start_money) / games
print("average win/lost per game: ", curr_win)
past_win = -1000
while(curr_win > past_win):
past_win = curr_win
markov.update_probs()
start_money = 0
end_money = start_money
policy_params = {'actions': markov.actions,
'actions_prob': markov.actions_probs,
'threshold': 90}
for games in range(100000):
deck.start_game()
final_hand, d_hand, rewards = play_game(
deck, [1], probability_based_policy, **policy_params)
end_money = end_money + rewards[0]
markov.add_game(final_hand[0], rewards[0])
curr_win = (end_money - start_money) / games
print("average win/lost per game: ", curr_win)
print(markov.actions_probs)
def test_game():
b = BjDeck(6)
policy_params = {'policy': WEB_POLICY}
a, b, c = play_game(b, [100], fixed_policy, **policy_params)
print(a, b, c)