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 test_reverse(self):
board = Board.init_board()
actual = put_and_reverse((4, 5), board)
expected = Board.init_board().board
expected[4][4] = 1
expected[4][5] = 1
np.testing.assert_array_equal(expected, actual.board)
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 test_multi_size_reverse(self):
board = Board.init_board()
board.board[4][5] = -1
actual = put_and_reverse((4, 6), board)
expected = Board.init_board().board
expected[4][4] = 1
expected[4][5] = 1
expected[4][6] = 1
np.testing.assert_array_equal(expected, actual.board)
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 test_index_array_of_bound(self):
board = Board(np.zeros((8, 8)))
board.board[0][0] = -1
board.board[0][2] = -1
board.board[0][3] = 1
actual = put_and_reverse((0, 1), board)
expected = np.zeros_like(board.board)
expected[0][0] = -1
expected[0][1] = 1
expected[0][2] = 1
expected[0][3] = 1
np.testing.assert_array_equal(expected, actual.board)
def _put_and_reverse(self, hand: Union[Hand, Tuple[int, int]], board: Board) -> Board:
return put_and_reverse(hand, board)