def minimax(tabuleiro, depth, alpha, beta, maximizingPlayer, current_move, player):
best_move = current_move
score = tabuleiro.calculate_board_score(player)
if depth == 0 or score == inf:
return score, current_move
possible_moves = utils.get_possible_moves(tabuleiro.get_board())
if maximizingPlayer:
maxEval = -inf
for move in possible_moves:
evalue, pos = minimax(utils.make_move(copy.deepcopy(tabuleiro), move, 1), depth-1, alpha, beta, False, move, player)
if maxEval < evalue:
maxEval = evalue
best_move = move
alpha = max(alpha, evalue)
if alpha >= beta:
break
return maxEval, best_move
else:
minEval = inf
for move in possible_moves:
evalue, pos = minimax(utils.make_move(copy.deepcopy(tabuleiro), move, 2), depth - 1, alpha, beta, True, move, player)
if minEval > evalue:
minEval = evalue
best_move = move
beta = min(beta, evalue)
if alpha >= beta:
break
return minEval, best_move
def move(self, board) -> Tuple[int, int, Action]:
if random() < self.epsilon:
possible_moves = get_possible_moves(board, self.mark)
if possible_moves:
chosen_move = choice(possible_moves)
move_hash = ALL_MOVES.index(chosen_move)
else:
move_hash = 0
chosen_move = ALL_MOVES[move_hash]
else:
predictions = self.model.predict(encode_board(board, self.mark))[0]
possible_moves_hashes = get_encoded_possible_moves(
board, self.mark)
move_hash = next((x for _, x in sort_predictions(predictions)
if x in possible_moves_hashes), 0)
chosen_move = ALL_MOVES[move_hash]
if self.training:
self.tau += 1
if self.tau > self.max_tau:
self.tau = 0
self.update_target_model()
self.memorize(board, move_hash, 0, False)
return chosen_move
def move(self, board) -> Tuple[int, int, Action]:
mark = self.mark
if self.training:
move_hash = 0
chosen_move = ALL_MOVES[move_hash]
if random() < self.epsilon:
possible_moves = get_possible_moves(board, mark)
if possible_moves:
chosen_move = choice(possible_moves)
move_hash = ALL_MOVES.index(chosen_move)
else:
predictions = self.model.predict(encode_board(board, mark))[0]
possible_moves_hashes = get_encoded_possible_moves(board, mark)
move_hash = next((x for _, x in sort_predictions(predictions)
if x in possible_moves_hashes), 0)
chosen_move = ALL_MOVES[move_hash]
self.buffer.append((board, move_hash))
return chosen_move
else:
predictions = self.model.predict(encode_board(board, mark))[0]
possible_moves_hashes = get_encoded_possible_moves(board, mark)
move_hash = next((x for _, x in sort_predictions(predictions)
if x in possible_moves_hashes), 0)
return ALL_MOVES[move_hash]
def minimax(game_state, turn, maximizer_turn, round_num=0, top=False):
node_key = hash_game_state(game_state)
if node_key in MINIMAX_TREE:
return MINIMAX_TREE[node_key]
possible_moves = get_possible_moves(game_state)
if not possible_moves or get_winner(game_state)[0] is not None:
return _score(game_state, maximizer_turn), None, round_num
if turn == maximizer_turn:
best = -2 # Lower than lowest possible score.
best_move = None
best_move_depth = 10 # Larger than deepest possible tree.
for i, move in enumerate(possible_moves):
move_score, _, move_depth = minimax(
_simulate_move(game_state, move, turn),
toggle_turn(turn),
maximizer_turn,
round_num + 1
)
# break ties by preferring faster wins
is_faster = move_depth < best_move_depth
if move_score > best or (move_score == best and is_faster):
best_move = move
best = move_score
best_move_depth = move_depth
# Print a progress bar at the top level
if top:
progress = 100. * (i + 1.) / len(possible_moves)
sys.stdout.write(
'Computer is thinking... Progress: %.02f%%\r' % progress)
sys.stdout.flush()
ret = (best, best_move, best_move_depth)
MINIMAX_TREE[node_key] = ret
return ret
else:
best = 2 # Higher than highest possible score
best_move = None
best_move_depth = 10 # Larger than deepest possible tree.
for move in possible_moves:
move_score, _, move_depth = minimax(
_simulate_move(game_state, move, turn),
toggle_turn(turn),
maximizer_turn,
round_num + 1
)
# break ties by preferring faster wins
is_faster = move_depth < best_move_depth
if move_score < best or (move_score == best and is_faster):
best_move = move
best = move_score
best_move_depth = move_depth
ret = (best, best_move, best_move_depth)
MINIMAX_TREE[node_key] = ret
return ret
def get_ordered_next_games(self, mark: Mark, alpha: float, beta: float,
desc: bool = True):
possible_games = [(move, self.move(*move, mark))
for move in get_possible_moves(self.board, mark)]
# add some random noise to randomize results
possible_games.sort(key=lambda x: (x[1].maybe_evaluate(mark, 0, alpha, beta, 0),
random.random()),
reverse=desc)
return possible_games
def random(game_state, turn):
'''Random Algorithm'''
possible_moves = get_possible_moves(game_state)
return choice(possible_moves)
def move(self, board) -> Tuple[int, int, Action]:
possible_moves = get_possible_moves(board, self.mark)
return choice(possible_moves) if possible_moves else ALL_MOVES[0]
def move(self, board) -> Tuple[int, int, Action]:
possible_moves = get_possible_moves(board, self.mark)
possible_moves.sort(key=lambda m: (evaluate_move(
board, self.mark, m, self.eval_rows, self.eval_cols, self.
eval_diagonals), random.random()))
return possible_moves[-1]
# -*- coding: utf-8 -*-
import board
import utils
from math import inf
import minimax
tabuleiro = board.Board()
lista = tabuleiro.calculate_board_score(1)
#print(lista)
moves = utils.get_possible_moves(tabuleiro.get_board())
print(moves)
score, move = minimax.minimax(tabuleiro, 3, -inf, inf, True, [0, 0], 1)
print("SCORE: ")
print(score)
print("Movimento")
print(move)
