def __init__(self,
side,
network,
debugging=True,
n_playout=800,
search_threads=16,
virtual_loss=0.02,
policy_loop_arg=True,
c_puct=5,
dnoise=False,
temp_round=conf.train_temp_round,
can_surrender=False,
surrender_threshold=-0.99,
allow_legacy=False,
repeat_noise=True):
super(NetworkPlayer, self).__init__(side)
self.network = network
self.debugging = debugging
self.queue = Queue(400)
self.temp_round = temp_round
self.can_surrender = can_surrender
self.allow_legacy = allow_legacy
self.surrender_threshold = surrender_threshold
self.repeat_noise = repeat_noise
self.mcts_policy = mcts_async.MCTS(self.policy_value_fn_queue,
n_playout=n_playout,
search_threads=search_threads,
virtual_loss=virtual_loss,
policy_loop_arg=policy_loop_arg,
c_puct=c_puct,
dnoise=dnoise)
if move in legal_move_b:
move = board.flipped_uci_labels([move])[0]
action_probs.append((move,prob))
else:
for move,prob in zip(uci_labels,policyout):
if move in legal_move:
action_probs.append((move,prob))
action_probs = sorted(action_probs,key=lambda x:x[1])
return action_probs, valout
for one_play in range(10):
states = []
moves = []
game_states = GameState()
mcts_policy_w = mcts_async.MCTS(policy_value_fn_async_batch,n_playout=1600,search_threads=32,virtual_loss=0.03)
mcts_policy_b = mcts_async.MCTS(policy_value_fn_async_batch,n_playout=1600,search_threads=32,virtual_loss=0.03)
result = 'peace'
for i in range(150):
begin = time.time()
is_end,winner = game_states.game_end()
if is_end == True:
result = winner
break
start = time.time()
if i % 2 == 0:
player = 'w'
if i < 18:
temp = 1
else:
with graph.as_default():
saver = tf.train.Saver(var_list=tf.global_variables())
saver.restore(sess, model_dir)
print('param updated {}'.format(model_dir))
nm.updated(latest_model_name)
print("current weight {}".format(nm.netname))
states = []
moves = []
game_states = GameState()
mcts_policy_w = mcts_async.MCTS(policy_value_fn_queue,
n_playout=800,
search_threads=16,
virtual_loss=0.02,
policy_loop_arg=True,
c_puct=5,
dnoise=False)
mcts_policy_b = mcts_async.MCTS(policy_value_fn_queue,
n_playout=800,
search_threads=16,
virtual_loss=0.02,
policy_loop_arg=True,
c_puct=5,
dnoise=False)
result = 'peace'
peace_round = 0
remain_piece = countpiece(game_states.statestr)
can_surrender = random.random() > 0.1