def test_fight2(self):
player_blue = NNPlayer(Color.BLUE,
n_simulations=100,
janggi_net=JanggiNetwork(),
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
player_red = NNPlayer(Color.RED,
n_simulations=100,
janggi_net=JanggiNetwork(),
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
fight(player_blue, player_red, 100)
def test_mcts_vs_random(self):
player_blue = RandomPlayer(Color.BLUE)
player_red = RandomMCTSPlayer(Color.RED,
n_simulations=200,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
winner = fight(player_blue, player_red, 200)
self.assertEqual(winner, Color.RED)
def organize_fight(self):
player_red = RandomPlayer(Color.RED)
player_blue = NNPlayer(Color.BLUE,
n_simulations=self.n_simulations,
janggi_net=self.predictor,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
fight(player_blue, player_red, self.iter_max)
player_red = RandomMCTSPlayer(
Color.RED,
n_simulations=self.n_simulations_opponent,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
player_blue = NNPlayer(Color.BLUE,
n_simulations=self.n_simulations,
janggi_net=self.predictor,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
fight(player_blue, player_red, self.iter_max)
result = load()
while True:
players = ["random_mcts"]
model_saver = ModelSaver()
for i in range(model_saver.get_last_weight_index() + 1):
players.append(i)
if len(players) == 1:
print("Not enough players")
continue
player_one_name, player_two_name = random.sample(players, k=2)
player_one = get_player(player_one_name, Color.BLUE, model_saver)
player_two = get_player(player_two_name, Color.RED, model_saver)
player_one_name = str(player_one_name)
player_two_name = str(player_two_name)
if player_one_name not in result:
result[player_one_name] = dict()
if player_two_name not in result[player_one_name]:
result[player_one_name][player_two_name] = [0, 0]
winner = fight(player_one, player_two, 200)
if winner == Color.BLUE:
result[player_one_name][player_two_name][0] += 1
else:
result[player_one_name][player_two_name][1] += 1
print_results(result)
save(result)
from ia.random_mcts_player import NNPlayer, fight, RandomMCTSPlayer
from janggi.parameters import N_ITERATIONS, DEFAULT_N_SIMULATIONS
from janggi.utils import Color
# Example of command:
# python3 show_match_nn.py --number_simulations 800 --n_iterations 200 --root_file_inference /tmp/showmatch --parallel_mcts True --n_threads_mcts 10
player_blue = NNPlayer(Color.BLUE,
n_simulations=DEFAULT_N_SIMULATIONS,
janggi_net=FilePredictor(),
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01,
print_info=True)
player_red = RandomMCTSPlayer(Color.RED,
n_simulations=16000,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01,
print_info=True)
# player_red = NNPlayer(Color.RED, n_simulations=DEFAULT_N_SIMULATIONS,
# janggi_net=FilePredictor(),
# temperature_start=0.01,
# temperature_threshold=30,
# temperature_end=0.01,
# print_info=True)
fight(player_blue, player_red, N_ITERATIONS, print_board=True)
from ia.random_mcts_player import fight, RandomMCTSPlayer
from janggi.action import Action
from janggi.player import Player
from janggi.utils import Color
class HumanPlayer(Player):
def play_action(self):
while True:
read_data = input("Enter your action:")
try:
action = Action.from_uci_usi(read_data.strip())
break
except IndexError:
print("Invalid Action")
return action
if __name__ == "__main__":
player_blue = HumanPlayer(Color.BLUE)
player_red = RandomMCTSPlayer(Color.RED,
n_simulations=16000,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01,
think_when_other=True,
print_info=True)
winner = fight(player_blue, player_red, 200, print_board=True)
def continuous_learning_once(self):
# First, train
for _ in range(EPOCH_NUMBER_CONTINUOUS):
training_set = []
for example in _raw_to_examples(
self.model_saver.all_episodes_raw_iterators(),
PROP_POPULATION_FOR_LEARNING):
training_set.append(example)
if len(training_set) > N_LAST_GAME_TO_CONSIDER:
if not TRAIN_ON_ALL:
break
self.train(training_set)
training_set = []
self.train(training_set)
# Then, fight!
# old_model = copy.deepcopy(self.predictor)
self.model_saver.load_latest_model(self.old_model, None)
self.old_model.to(DEVICE)
victories = 0
print("Start the fights!")
for i in range(N_FIGHTS):
if i < N_FIGHTS / 2:
print("I am BLUE")
new_player = NNPlayer(Color.BLUE,
n_simulations=self.n_simulations,
janggi_net=self.predictor,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
old_player = NNPlayer(Color.RED,
n_simulations=self.n_simulations,
janggi_net=self.old_model,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
winner = fight(new_player, old_player, self.iter_max)
if winner == Color.BLUE:
victories += 1
else:
print("I am RED")
new_player = NNPlayer(Color.RED,
n_simulations=self.n_simulations,
janggi_net=self.predictor,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
old_player = NNPlayer(Color.BLUE,
n_simulations=self.n_simulations,
janggi_net=self.old_model,
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
winner = fight(old_player, new_player, self.iter_max)
if winner == Color.RED:
victories += 1
if (victories + N_FIGHTS - i -
1) / N_FIGHTS * 100 < VICTORY_THRESHOLD:
# There is no more hope...
break
victory_percentage = victories / N_FIGHTS * 100
if victory_percentage > VICTORY_THRESHOLD:
# Replace model
print("The model was good enough", victory_percentage)
self.model_saver.save_weights(self.predictor,
optimizer=self.optimizer)
else:
# We do not save the model
print("The model was not good enough", victory_percentage)
