def test_available_moves(self):
size = 3
board = clean_board(size)
self.assertEqual(len(available_moves(board)), 9)
apply_move_inplace(board, (0, 0), 1)
self.assertEqual(len(available_moves(board)), 8)
board = np.ones((size, size), dtype=np.int)
self.assertEqual(len(available_moves(board)), 0)
def add_board(self, board):
board_hash = hash_board(board)
if self.q_table.get(board_hash) is None:
legal_moves = available_moves(board)
self.q_table[board_hash] = {move: 1.0 for move in legal_moves}
return board_hash
def get_move(self, board):
legal_moves = available_moves(board)
new_boards = np.array([apply_move(board, move, self.side_to_play) for move in legal_moves])
# print(new_boards.reshape(len(legal_moves), 9))
# possible_boards = [apply_move(board, move, self.side_to_play) for move in legal_moves]
evaluations = self.model.predict(new_boards.reshape(len(legal_moves), 9))
print(legal_moves)
print(evaluations)
return legal_moves[self.min_max_best_move(evaluations)]
def learn_q(self, board, move):
board_hash = self.add_board(board)
new_board = apply_move(board, move, self.side)
new_board_hash = self.add_board(new_board)
reward = self.calculate_reward(new_board)
if reward != 0 or len(available_moves(new_board_hash)) == 0:
expected = reward
else:
expected_rewards = self.q_table[new_board_hash]
expected = reward + (0.9 * max(expected_rewards.values()))
change = 0.3 * (expected - self.q_table[board_hash][move])
self.q_table[board_hash][move] += change
def get_move(self, board, side):
return random.choice(available_moves(board))