def get_experiment_environment(**args):
from utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
tf_context = setup_tensorflow_session()
return tf_context
def get_experiment_environment(**args):
from utils import setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * 0
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
logger_context = logger.scoped_configure(
dir=None, format_strs=['stdout', 'log', 'csv'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def get_experiment_environment(**args):
from utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
logger_context = logger.scoped_configure(dir=args['env'],
format_strs=['stdout', 'log',
'csv'] if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def get_experiment_environment(**args):
process_seed = 1234 + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(1234)
setup_mpi_gpus()
logger_context = logger.scoped_configure(
dir='C:/Users/Elias/Desktop/savedunc/' + MODE + '_' +
datetime.now().strftime('%Y_%m_%d_%H_%M_%S'),
format_strs=['stdout', 'log', 'csv', 'tensorboard']
if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def get_experiment_environment(**args):
from utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
logdir = './' + args["logdir"] + '/' + datetime.datetime.now().strftime(
args["expID"] + "-openai-%Y-%m-%d-%H-%M-%S-%f")
logger_context = logger.scoped_configure(
dir=logdir,
format_strs=['stdout', 'log', 'csv', 'tensorboard']
if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context, logdir
def get_experiment_environment(**args):
log_directory = osp.join(
'./output/' +
datetime.datetime.now().strftime("openai-%Y-%m-%d-%H-%M-%S-%f"))
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
# TODO: turn this back up when have GPU support!
# setup_mpi_gpus()
logger_context = logger.scoped_configure(
dir=log_directory,
format_strs=['stdout', 'log', 'csv']
if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def play_experiment(**args):
# 该函数在本文件中定义. 对不同的环境进行各种 wrapper. 这是一个函数, 并没有真正初始化环境
make_env = partial(make_env_all_params, args=args)
# 初始化
tester = Tester(make_env=make_env,
num_timesteps=args['num_timesteps'],
hps=args,
envs_per_process=1)
from utils import setup_tensorflow_session
tf_sess = setup_tensorflow_session()
saver = tf.train.Saver()
# model_path
model_path = args['model_path']
with tf_sess:
tester.play(tf_sess, args, saver, model_path)
def get_experiment_environment(**args):
from utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
time = datetime.datetime.now().strftime("%m-%d-%H-%M-%S")
path_with_args = './logs/' + '_'.join([
time, args['exp_name'], args['env_kind'], args['feature_space'],
str(args['envs_per_process']),
str(args['train_discriminator']),
str(args['discriminator_weighted'])
])
format_strs = ['stdout', 'log', 'csv', 'tensorboard'
] if MPI.COMM_WORLD.Get_rank() == 0 else ['log']
logger_context = logger.scoped_configure(dir=path_with_args,
format_strs=format_strs)
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def get_experiment_environment(**args):
# 初始化 MPI 相关的量
from utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
logger_dir = './logs/' + datetime.datetime.now().strftime(
args["env"] + "-" + args["reward_type"] + "-" +
str(args["nepochs_dvae"]) + "-" + str(args["stickyAtari"]) +
"-%Y-%m-%d-%H-%M-%S-%f")
logger_context = logger.scoped_configure(
dir=logger_dir,
format_strs=['stdout', 'log', 'csv']
if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
# bai. 新增 saver 用于保存权重
saver = tf.train.Saver()
return logger_context, tf_context, saver, logger_dir
def start_experiment(**args):
# create environment
# coinrun environment is already vectorized
env, test_env = make_env_all_params(args=args)
# set random seeds for reproducibility
utils.set_global_seeds(seed=args['seed'])
# create tf.session
tf_sess = utils.setup_tensorflow_session()
if args['server_type'] == 'local':
logger_context = logger.scoped_configure(dir=args['log_dir'],
format_strs=['stdout', 'csv'])
else:
logger_context = logger.scoped_configure(dir=args['log_dir'],
format_strs=['csv'])
with logger_context, tf_sess:
print("logging directory: {}".format(args['log_dir']))
# create trainer
trainer = Trainer(env=env, test_env=test_env, args=args)
if args['evaluation'] == 1:
# load_path is changed to model_path
print('run.py, def start_experiment, evaluating model: {}'.format(
args['load_path']))
trainer.eval()
# this is for visualizing the loss landscape
elif args['visualize'] == 1:
print('running visualization...')
trainer.visualize()
else:
print('run.py, def start_experiment, training begins...')
trainer.train()
