def play(network, ai_param=None):
if ai_param is None:
ai_param = dict(play_count=100)
board = Board.init_board()
ai1 = AlphaZero(network, history=True, **ai_param)
ai2 = AlphaZero(network, history=True, **ai_param)
hands = []
while True:
if not extract_valid_hand(board):
board.side ^= True
hands.append(Hand.pass_hand())
if not extract_valid_hand(board):
break
if board.side:
hand = ai1.put(board, hands)
else:
hand = ai2.put(board, hands)
hands.append(hand)
board = put_and_reverse(hand, board)
result = judge_simple(board)
history1 = ai1.history if isinstance(ai1, AlphaZero) else []
history2 = ai2.history if isinstance(ai1, AlphaZero) else []
for x in history1:
x[-1] = result
for x in history2:
x[0] = x[0] * -1
x[-1] = -result
return history1 + history2
def main(sente=True):
board = Board.init_board()
ai = MiniMaxAI(5)
print(view_board(board))
hands = []
while True:
if not extract_valid_hand(board):
board.side ^= True
hands.append(Hand.pass_hand())
if not extract_valid_hand(board):
break
if board.side ^ sente:
hand = input_hand(board)
else:
hand = ai.put(board, hands)
hands.append(hand)
print("AI put: {}".format(hand))
if (board.board == 0).sum() < 12:
# 計算時間に余裕があるのでdeepに読む
ai.depth = 8
board = put_and_reverse(hand, board)
print(view_board(board))
print("=" * 10 + " GAME OVER " + "=" * 10)
x_count = (board.board == 1).sum()
o_count = (board.board == -1).sum()
print("x: {}, o: {}".format(x_count, o_count))
def main(sente=True):
board = Board.init_board()
board.side = sente
ai = MonteCarloAI()
print(view_board(board))
while True:
if not extract_valid_hand(board):
board.side ^= True
if not extract_valid_hand(board):
break
if board.side:
hand = input_hand(board)
else:
if (board.board == 0).sum() < 8:
# 計算時間に余裕があるので全探索
hand = ai.put_exhaustive_search(board)
else:
hand = ai.put(board)
print("AI put: {}".format(hand))
board = put_and_reverse(hand, board)
print(view_board(board))
print("=" * 10 + " GAME OVER " + "=" * 10)
x_count = (board.board == 1).sum()
o_count = (board.board == -1).sum()
print("x: {}, o: {}".format(x_count, o_count))
def evaluate(ai1, ai2, sente=True, is_view=False):
board = Board.init_board()
if is_view:
print(view_board(board))
hands = []
while True:
if not extract_valid_hand(board):
board.side ^= True
hands.append(Hand.pass_hand())
# print("pass hand!!")
if not extract_valid_hand(board):
break
if board.side is sente:
hand = ai1.put(board, hands)
else:
hand = ai2.put(board, hands)
hands.append(hand)
board = put_and_reverse(hand, board)
if is_view:
print(view_board(board))
# print(ai2.history)
if is_view:
print("=" * 10 + " GAME OVER " + "=" * 10)
x_count = (board.board == 1).sum()
o_count = (board.board == -1).sum()
print("x: {}, o: {}".format(x_count, o_count))
return judge_simple(board) * (1 if sente else -1)
def _calc_valid_hand_p(self, p: np.ndarray,
board: Board) -> Tuple[np.ndarray, List[Hand]]:
p = p[:-1].reshape((8, 8))
hands = extract_valid_hand(board)
valid_p = np.array([p[hand.hand[0]][hand.hand[1]] for hand in hands])
result = valid_p / sum(valid_p + 1e-18)
return result, hands
def play_out(board: Board) -> float:
root_board = board
while not is_finished(board):
valid_hands = extract_valid_hand(board)
if len(valid_hands) == 0:
board = Board(board.board, not board.side)
continue
hand = np.random.choice(valid_hands)
board = put_and_reverse(hand, board)
return judge_simple(board) * (1 if root_board.side else -1)
def _extract_valid_hand(self, board: Board):
ret = extract_valid_hand(board)
if ret:
return ret
else:
return [Hand.pass_hand()]
def put(self, board: Board, hands: List[Hand]) -> Hand:
hands = extract_valid_hand(board)
return np.random.choice(hands)
def test_false_side(self):
board = Board.init_board(False)
actual = extract_valid_hand(board)
expected = [(2, 4), (4, 2), (3, 5), (5, 3)]
self.assertEquals(set(expected), set(i.hand for i in actual))