def experience_simulation(num_games, agent1, agent2):
collector1 = rl.ExperienceCollector()
collector2 = rl.ExperienceCollector()
color1 = Player.black
for i in range(num_games):
collector1.begin_episode()
agent1.set_collector(collector1)
collector2.begin_episode()
agent2.set_collector(collector2)
if color1 == Player.black:
black_player, white_player = agent1, agent2
else:
white_player, black_player = agent2, agent1
game_record = simulate_game(black_player, white_player)
if game_record.winner == color1:
collector1.complete_episode(reward=1)
collector2.complete_episode(reward=-1)
else:
collector2.complete_episode(reward=1)
collector1.complete_episode(reward=-1)
color1 = color1.other
return rl.combine_experience([collector1, collector2])
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--board-size', type=int, default=19)
parser.add_argument('--learning-agent', required=True)
parser.add_argument('--num-games', '-n', type=int, default=10)
# parser.add_argument('--game-log-out', required=True)
parser.add_argument('--experience-out', required=True)
parser.add_argument('--temperature', type=float, default=0.0)
args = parser.parse_args()
global BOARD_SIZE
BOARD_SIZE = args.board_size
agent1 = rl.load_ac_agent(h5py.File(args.learning_agent))
agent2 = rl.load_ac_agent(h5py.File(args.learning_agent))
agent1.set_temperature(args.temperature)
agent2.set_temperature(args.temperature)
collector1 = rl.ExperienceCollector()
collector2 = rl.ExperienceCollector()
color1 = Player.black
# logf = open(args.game_log_out, 'a')
# logf.write('Begin training at %s\n' % (datetime.datetime.now().strftime('%Y-%m-%d %H:%M'),))
for i in range(args.num_games):
print('Simulating game %d/%d...' % (i + 1, args.num_games))
collector1.begin_episode()
agent1.set_collector(collector1)
collector2.begin_episode()
agent2.set_collector(collector2)
if color1 == Player.black:
black_player, white_player = agent1, agent2
else:
white_player, black_player = agent1, agent2
game_record = simulate_game(black_player, white_player)
if game_record.winner == color1:
print('Agent 1 wins.')
collector1.complete_episode(reward=1)
collector2.complete_episode(reward=-1)
else:
print('Agent 2 wins.')
collector2.complete_episode(reward=1)
collector1.complete_episode(reward=-1)
color1 = color1.other
experience = rl.combine_experience([collector1, collector2])
# logf.write('Saving experience buffer to %s\n' % args.experience_out)
with h5py.File(args.experience_out, 'w') as experience_outf:
experience.serialize(experience_outf)
def do_self_play(board_size, agent1_filename, agent2_filename, num_games,
experience_filename, gpu_frac):
kerasutil.set_gpu_memory_target(gpu_frac)
random.seed(int(time.time()) + os.getpid())
np.random.seed(int(time.time()) + os.getpid())
agent1 = load_agent(agent1_filename)
agent2 = load_agent(agent2_filename)
collector1 = rl.ExperienceCollector()
color1 = Player.black
for i in range(num_games):
print('Simulating game %d/%d...' % (i + 1, num_games))
collector1.begin_episode()
agent1.set_collector(collector1)
if color1 == Player.black:
black_player, white_player = agent1, agent2
else:
white_player, black_player = agent1, agent2
game_record = simulate_game(black_player, white_player, board_size)
if game_record.winner == color1:
print('Agent 1 wins.')
collector1.complete_episode(reward=1)
else:
print('Agent 2 wins.')
collector1.complete_episode(reward=-1)
color1 = color1.other
experience = rl.combine_experience([collector1])
print('Saving experience buffer to %s\n' % experience_filename)
with h5py.File(experience_filename, 'w') as experience_outf:
experience.serialize(experience_outf)
def do_self_play(board_size,
agent1_filename,
agent2_filename,
num_games,
temperature,
experience_filename,
chunk=100,
gpu_frac=0.95):
kerasutil.set_gpu_memory_target(gpu_frac)
random.seed(int(time.time()) + os.getpid())
np.random.seed(int(time.time()) + os.getpid())
agent1 = load_agent(agent1_filename)
agent1.set_temperature(temperature)
agent1.set_policy('eps-greedy')
agent2 = load_agent(agent2_filename)
agent2.set_temperature(temperature)
agent2.set_policy('eps-greedy')
color1 = Player.black
times = int(num_games / chunk)
for current_chunk in range(times):
print('Текущая порция %d' % current_chunk)
collector1 = rl.ExperienceCollector()
for i in range(chunk):
print('Симуляция игры %d/%d...' % (i + 1, chunk))
collector1.begin_episode()
agent1.set_collector(collector1)
if color1 == Player.black:
black_player, white_player = agent1, agent2
else:
white_player, black_player = agent1, agent2
game_record = simulate_game(black_player, white_player, board_size)
cnt_moves = len(game_record.moves)
if game_record.winner == color1:
print('Агент 1 выигрывает, время: %s' %
(datetime.datetime.now()))
collector1.complete_episode(reward=1)
else:
print('Агент 2 выигрывает, время: %s' %
(datetime.datetime.now()))
collector1.complete_episode(reward=-1)
print('Количество ходов в игре = ', cnt_moves)
color1 = color1.other
experience = rl.combine_experience([collector1])
print('Saving experience buffer to %s\n' % (experience_filename + str(
(current_chunk + 1) * chunk) + '.h5'))
with h5py.File(
experience_filename + '_' + str(
(current_chunk + 1) * chunk) + '.h5',
'w') as experience_outf:
experience.serialize(experience_outf)
def main():
# parser = argparse.ArgumentParser()
# parser.add_argument('--learning-agent', required=True)
# parser.add_argument('--num-games', '-n', type=int, default=10)
# parser.add_argument('--game-log-out', required=True)
# parser.add_argument('--experience-out', required=True)
# parser.add_argument('--temperature', type=float, default=0.0)
#
# args = parser.parse_args()
# 9.19
agent1 = agent.load_policy_agent(h5py.File(agent_filename))
agent2 = agent.load_policy_agent(h5py.File(agent_filename))
collector1 = rl.ExperienceCollector()
collector2 = rl.ExperienceCollector()
agent1.set_collector(collector1)
agent2.set_collector(collector2)
# color1 = Player.black
# logf = open(args.game_log_out, 'a')
# logf.write('Begin training at %s\n' % (datetime.datetime.now().strftime('%Y-%m-%d %H:%M'),))
# 9.20
for i in range(num_games):
collector1.begin_episode()
collector2.begin_episode()
game_record = simulate_game(agent1, agent2)
if game_record.winner == Player.black:
collector1.complete_episode(reward=1) # Agent1 won the game, so +reward
collector2.complete_episode(reward=-1)
else:
collector2.complete_episode(reward=1) # Agent2 won the game
collector1.complete_episode(reward=-1)
experience = rl.combine_experience([collector1, collector2])
# logf.write('Saving experience buffer to %s\n' % args.experience_out)
with h5py.File(experience_filename, 'w') as experience_outf:
experience.serialize(experience_outf)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--board-size', type=int, required=True)
parser.add_argument('--learning-agent', required=True)
parser.add_argument('--num-games', '-n', type=int, default=10)
parser.add_argument('--experience-out', required=True)
args = parser.parse_args()
agent_filename = args.learning_agent
experience_filename = args.experience_out
num_games = args.num_games
global BOARD_SIZE
BOARD_SIZE = args.board_size
agent1 = agent.load_policy_agent(h5py.File(agent_filename))
agent2 = agent.load_policy_agent(h5py.File(agent_filename))
collector1 = rl.ExperienceCollector()
collector2 = rl.ExperienceCollector()
agent1.set_collector(collector1)
agent2.set_collector(collector2)
for i in range(num_games):
collector1.begin_episode()
collector2.begin_episode()
game_record = simulate_game(agent1, agent2)
if game_record.winner == Player.black:
collector1.complete_episode(reward=1)
collector2.complete_episode(reward=-1)
else:
collector2.complete_episode(reward=1)
collector1.complete_episode(reward=-1)
experience = rl.combine_experience([collector1, collector2])
with h5py.File(experience_filename, 'w') as experience_outf:
experience.serialize(experience_outf)
def do_self_play(board_size, agent1_filename, agent2_filename, num_games,
temperature, experience_filename, gpu_frac):
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
# Restrict TensorFlow to only use the first GPU
try:
tf.config.experimental.set_visible_devices(gpus[0], 'GPU')
tf.config.experimental.set_memory_growth(gpus[0], True)
except RuntimeError as e:
print(e)
random.seed(int(time.time()) + os.getpid())
np.random.seed(int(time.time()) + os.getpid())
with h5py.File(agent1_filename, 'r') as agent1f:
agent1 = agent.load_policy_agent(agent1f)
agent1.set_temperature(temperature)
with h5py.File(agent2_filename, 'r') as agent2f:
agent2 = agent.load_policy_agent(agent2f)
collector1 = rl.ExperienceCollector()
color1 = Player.black
for i in range(num_games):
print('Simulating game %d/%d...' % (i + 1, num_games))
collector1.begin_episode()
agent1.set_collector(collector1)
if color1 == Player.black:
black_player, white_player = agent1, agent2
else:
white_player, black_player = agent1, agent2
game_record = simulate_game(black_player, white_player, board_size)
if game_record.winner == color1:
print('Agent 1 wins.')
collector1.complete_episode(reward=1)
else:
print('Agent 2 wins.')
collector1.complete_episode(reward=-1)
color1 = color1.other
experience = rl.combine_experience([collector1])
print('Saving experience buffer to %s\n' % experience_filename)
with h5py.File(experience_filename, 'w') as experience_outf:
experience.serialize(experience_outf)
def main():
learning_agent = input("Бот: ")
temperature = float(input('Температура = '))
game_log = input('game_log: ')
experience_out = input('experience_out: ')
num_games = int(input('Количество игр = '))
try:
chunk_size = int(input('Количество игр в "порции" ='))
except:
chunk_size = 100
pth = '//media//nail//SSD_Disk//Experience//'
learning_agent = '//media//nail//SSD_Disk//Models//' + learning_agent + '.h5'
game_log = pth + game_log + '_' + str(num_games)
experience_out = pth + experience_out + '_' + str(num_games) + '_' #+'.h5'
#args = parser.parse_args()
# ==================================================
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
import tensorflow as tf
config = tf.compat.v1.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.95
config.gpu_options.allow_growth = True
config.log_device_placement = True
sess = tf.compat.v1.Session(config=config)
tf.compat.v1.keras.backend.set_session(sess)
# ==================================================
agent1 = agent.load_policy_agent(h5py.File(learning_agent, "r"))
agent2 = agent.load_policy_agent(h5py.File(learning_agent, "r"))
agent1.set_temperature(temperature)
agent2.set_temperature(temperature)
k = 0
j = 0
for i in range(num_games + 1):
if j == 0:
game_log_out = game_log + '_' + str((k + 1) * chunk_size) + ".txt"
logf = open(game_log_out, 'a')
logf.write('Начало игр в %s\n' %
(datetime.datetime.now().strftime('%Y-%m-%d %H:%M'), ))
logf.write(
str((k + 1) * chunk_size) + ' из количества игр: ' +
str(num_games) + '\n')
print('Моделируемая игра %d/%d...' % (i + 1, num_games))
collector1 = rl.ExperienceCollector()
collector2 = rl.ExperienceCollector()
color1 = Player.black
j += 1
collector1.begin_episode()
agent1.set_collector(collector1)
collector2.begin_episode()
agent2.set_collector(collector2)
if color1 == Player.black:
black_player, white_player = agent1, agent2
else:
white_player, black_player = agent2, agent1
game_record = simulate_game(black_player, white_player)
print(" № игры : ", i + 1)
if game_record.winner == color1:
print('Агент 1 выигрывает.')
collector1.complete_episode(reward=1)
collector2.complete_episode(reward=-1)
else:
print('Агент 2 выигрывает.')
collector2.complete_episode(reward=1)
collector1.complete_episode(reward=-1)
color1 = color1.other
if i >= chunk_size and i % chunk_size == 0:
experience = rl.combine_experience([collector1, collector2])
experience_out_file = experience_out + str(
(k + 1) * chunk_size) + ".h5"
logf.write('Сохранение буфера в файл %s\n' % experience_out_file)
logf.write('Завершение игр %s\n' %
(datetime.datetime.now().strftime('%Y-%m-%d %H:%M'), ))
logf.close()
with h5py.File(experience_out_file, 'w') as experience_outf:
experience.serialize(experience_outf)
print('Записано игр: ', (k + 1) * chunk_size, ' из ', num_games,
' игр.')
k += 1
j = 0
