def min_select(boards, model, history):
if len(boards) == 0:
boards.append(chess.Board())
curr_board = boards[-1]
possible = list(curr_board.legal_moves)
min_val = float('inf')
for move in possible:
t_boards = copy.deepcopy(boards)
t_curr = copy.copy(t_boards[-1])
t_curr.push(move)
if t_curr.result() == '1-0':
return move
t_boards.append(t_curr)
val = model.evaluate(
np.expand_dims(get_simple_input(t_boards, history), 0))
if val < min_val:
min_val = val
best_move = move
return best_move
def __init__(self, model, boards, history):
self.history = history
self.model = model
self.boards = boards
self.leading_edge = None
self.edges = []
self.value = model.evaluate(
np.expand_dims(get_simple_input(self.boards, self.history), 0))
def max_select(boards, model, history):
curr_board = boards[-1]
possible = list(curr_board.legal_moves)
max_val = -1 * float('inf')
for move in possible:
t_boards = copy.deepcopy(boards)
t_curr = copy.deepcopy(t_boards[-1])
t_curr = t_curr.push(move)
t_boards.append(t_curr)
val = model.evaluate(
np.expand_dims(get_simple_input(t_boards, history), 0))
if val > max_val:
max_val = val
best_move = move
return best_move
def evaluate(boards, model, history):
inputs = np.expand_dims(get_simple_input(boards, history), 0)
return model.evaluate(inputs)
'''
if self.check_win() is not None:
return []
possible = np.where(self.board == 0)
return [(possible[0][i], possible[1][i])
for i in range(len(possible[0]))]
if __name__ == '__main__':
game = TicTacToe()
boards = [copy.deepcopy(game)]
while game.check_win() is None:
possible_moves = game.legal_moves
print(game.board)
print(possible_moves)
move_ind = int(input('Select: '))
move = possible_moves[move_ind]
game = game.push(move)
boards.append(game)
for b in boards:
print(b.board)
for i in range(1, len(boards) + 1):
inp = get_simple_input(boards[:i], 2)
print(inp[:, :, 0])
print(inp[:, :, 1])
print()
def get_value(model, t_boards, history):
return model.evaluate(
np.expand_dims(get_simple_input(t_boards, history), 0))
def main(args):
model = DQN()
model.load(args.path)
engine = chess.uci.popen_engine(os.environ['SFSH'])
engine.uci()
board = chess.Board()
player = 1
boards = [copy.deepcopy(board)]
while not board.is_game_over(claim_draw=True):
print('CURRENT BOARD')
print(board)
if player == 1:
engine.position(board)
stk_move, _ = engine.go()
vals = []
board_moves = []
possible_moves = list(board.legal_moves)
for move in possible_moves:
t_boards = copy.deepcopy(boards)
t_curr = copy.copy(t_boards[-1])
t_curr.push(move)
t_boards.append(t_curr)
board_moves.append(t_curr)
vals.append(
model.evaluate(
np.expand_dims(
get_simple_input(t_boards, args.history), 0)))
if stk_move == move:
stk_val = vals[-1]
stk_board = t_curr
top_5 = heapq.nlargest(3, vals)
bottom_5 = heapq.nsmallest(3, vals)
print('Stockfish')
print(stk_board)
print(stk_val)
print('Top 5')
for v in top_5:
print(board_moves[vals.index(v)])
print(v)
print()
print()
print('Bottom 5')
for v in bottom_5:
print(board_moves[vals.index(v)])
print(v)
print()
print()
if args.max_select:
move = possible_moves[vals.index(max(vals))]
if args.min_select:
move = possible_moves[vals.index(min(vals))]
cont = input('Continue?')
else:
#move = random.choice(list(board.legal_moves))
engine.position(board)
move, _ = engine.go()
player = (player % 2) + 1
board.push(move)
boards.append(copy.deepcopy(board))
if len(boards) > 8:
boards.pop(0)
res = board.result()