if monte_carlo and (round_index <= 1): # exploring starts
player.next_action = np.random.choice(card_game.num_actions)
else:
player.next_action = player.policy[policy_index]
return play_action(card_showing, player)
q_learning = MonteCarloLearning(card_game.num_states, card_game.num_actions)
monte = Player()
opponent = Player()
for episode_index in xrange(NUM_GAMES_TO_PLAY):
deck.shuffle_deck()
monte.pick_up_cards(deck.deal_cards(card_game.hand_size))
opponent.pick_up_cards(deck.deal_cards(card_game.hand_size))
q_learning.clear_states_seen()
for round_index in xrange(card_game.num_rounds):
sum_of_cards = 0.
if round_index % 2 == 0:
sum_of_cards = take_turn(monte, round_index, sum_of_cards, monte_carlo=True)
sum_of_cards += take_turn(opponent, round_index, sum_of_cards)
else:
card_showing = take_turn(opponent, round_index, sum_of_cards)
sum_of_cards = take_turn(monte, round_index, sum_of_cards, monte_carlo=True)
if monte.last_card_played + opponent.last_card_played <= 1:
