def select_action_by_rule(state):
valid_hands = get_complete_hand(str(
state.curr_cards)) if state.prev_hand.size() == 0 else get_beat_hand(
str(state.prev_hand), str(state.curr_cards))
valid_hands = valid_hands.split(" ")
valid_hands = [
CardSet.parse_from("")
if valid_hand == "x" else CardSet.parse_from(valid_hand)
for valid_hand in valid_hands
]
# sorted by split card length
left_cards = [
CardSet.diff(state.curr_cards, valid_hand)
for valid_hand in valid_hands
]
split_lens = [
0 if left_card.size() == 0 else len(
get_simple_hand(str(left_card)).split(" "))
for left_card in left_cards
]
hands_with_lens = list(zip(valid_hands, split_lens))
min_split_len = min(hands_with_lens, key=lambda tpl: tpl[1])
min_split_len_hands = [
hand_with_len[0] for hand_with_len in hands_with_lens
if hand_with_len[1] == min_split_len[1]
]
if len(min_split_len_hands) == 1:
return min_split_len_hands[0]
# sorted by left card length
left_len = [(hand, 0 if CardSet.diff(state.curr_cards, hand).size() == 0
else len(str(CardSet.diff(state.curr_cards, hand))))
for hand in min_split_len_hands]
min_left_len = min(left_len, key=lambda tpl: tpl[1])
min_left_len_hands = [
hand for hand, length in left_len if length == min_left_len[1]
]
if len(min_left_len_hands) == 1:
return min_left_len_hands[0]
# sorted by hand rank
card_hands = [
CardHand.parse_from('' if str(hand) == 'x' else str(hand))
for hand in min_left_len_hands
]
sorted_by_rank = sorted(card_hands, key=cmp_to_key(card_hand_compare))
return sorted_by_rank[0].card_set
def expand(self):
if self.state.curr_cards.size() == 0 or self.state.next_cards.size(
) == 0:
return False
valid_hands = get_complete_hand(str(
self.state.curr_cards)) if self.state.prev_hand.size(
) == 0 else get_beat_hand(str(self.state.prev_hand),
str(self.state.curr_cards))
valid_hands = [
CardSet.parse_from("")
if valid_hand == "x" else CardSet.parse_from(valid_hand)
for valid_hand in valid_hands.split(" ")
]
next_states = [
MCTSState(1 - self.state.player, self.state.next_cards,
CardSet.diff(self.state.curr_cards, valid_hand),
valid_hand, False if valid_hand.nonempty() else True)
for valid_hand in valid_hands
]
self.children = [
MCTSNode(self, next_state) for next_state in next_states
]
pn_scores = self._get_pn_scores()
vn_scores = self._get_vn_scores()
for child, pn_score, vn_score in zip(self.children, pn_scores,
vn_scores):
child.P = pn_score
child.V = vn_score
return True
def rollout(state):
# valid_hand = select_action_by_rule(state)
valid_hand = select_action_by_policy(state)
# print "%s: %s" % ("dz" if state.player == 1 else "nm", str(valid_hand))
left_cards = CardSet.diff(state.curr_cards, valid_hand)
if left_cards.size() == 0:
# print "%s win" % ("dz" if state.player == 1 else "nm")
return 1.0 if state.player == 1 else -1.0
next_state = MCTSState(1 - state.player, state.next_cards, left_cards,
valid_hand,
False if state.prev_hand.nonempty() else True)
return MCTSNode.rollout(next_state)