def dfs(board):
mem = {hash_board(board)}
# count = 1
# max_goal = sum(board[0])
stack = [[board, []]]
while stack:
board, solution = stack.pop(-1)
# count += 1
# if count % 10000 == 0:
# print(count, max_goal, len(mem))
# print_board(board)
for b, move in list(valid_moves(board))[::-1]:
if hash_board(b) in mem:
continue
if sum(b[0]) == 230:
print(b[0])
return solution + [(move, b)]
mem.add(hash_board(b))
# max_goal = max(sum(b[0]), max_goal)
stack.append([b, solution + [(move, b)]])
def choose_turn(self, turns, board):
if hash_board(board.grid) in self.q_table and uniform(0, 1) > RANDOM_CHOICE_CHANCE:
sorted_actions = sorted(self.q_table[hash_board(
board.grid)].items(), key=operator.itemgetter(1))
return eval(sorted_actions.pop()[0])
elif uniform(0, 1) < NAIVE_GIVEN_RANDOM:
s_turns = sorted(turns, key=lambda x: abs(x[0][0] - x[1][0]))
s_turns = sorted(turns, key=len)
# print(turns)
if abs(s_turns[len(s_turns) - 1][0][0] - s_turns[len(s_turns) - 1][1][0]) == 1:
return choice(s_turns)
else:
return s_turns.pop()
else:
return choice(turns)
def play(self, board, silent=False):
if not silent:
print(board)
turns = board.get_all_legal_moves(board.grid)
t = self.choose_turn(turns, board)
if self.previous_state:
if hash_board(board.grid) in self.q_table:
# with open("q_update.txt", "a") as f:
# f.write("old_q " + str(self.q_table[self.previous_state][self.previous_action]))
# f.write("\nmax " + str(max([self.q_table[hash_board(board.grid)][a]
# for a in self.q_table[hash_board(board.grid)]])))
# f.write("\ndifference " + str(max([self.q_table[hash_board(board.grid)][a]
# for a in self.q_table[hash_board(board.grid)]])
# - self.q_table[self.previous_state][self.previous_action]))
self.q_table[self.previous_state][self.previous_action] += (
LEARNING_RATE *
((get_score(board.grid, board.turn) -
get_score(self.previous_grid, board.turn)) +
DISCOUNT * self.q_table[hash_board(board.grid)][str(t)] -
self.q_table[self.previous_state][self.previous_action]))
# f.write("\n" + str(self.q_table[self.previous_state][self.previous_action]))
# f.write("\n\n")
else:
self.q_table[self.previous_state][self.previous_action] += (
LEARNING_RATE *
(get_score(board.grid, board.turn) -
self.q_table[self.previous_state][self.previous_action]))
# sleep(1)
self.previous_grid = board.grid
self.previous_state = hash_board(board.grid)
self.previous_action = str(t)
if self.previous_state not in self.q_table:
action_dict = dict()
for turn in turns:
action_dict[str(turn)] = 0
self.q_table[self.previous_state] = action_dict
if not silent:
print("Q-learn moves " + " to ".join([translate(c)
for c in t]) + "\n")
for i in range(0, len(t) - 1): # loop is here to support double jumps
board.move(t[i], t[i + 1])