def main(args):
U.make_session(num_cpu=1).__enter__()
set_global_seeds(args.seed)
env = gym.make(args.env_id)
task_name = get_task_short_name(args)
logger.configure(dir='log_trpo_cartpole/%s' % task_name)
def policy_fn(name, ob_space, ac_space, reuse=False):
return build_policy(env, 'mlp', value_network='copy')
import logging
import os.path as osp
import bench
env = bench.Monitor(
env,
logger.get_dir() and osp.join(logger.get_dir(), "monitor.json"))
env.seed(args.seed)
gym.logger.setLevel(logging.WARN)
args.checkpoint_dir = osp.join(args.checkpoint_dir, task_name)
args.log_dir = osp.join(args.log_dir, task_name)
if args.task == 'train':
from utils.mujoco_dset import Dset_gym
expert_observations = np.genfromtxt(
'expert_data/cartpole/observations.csv')
expert_actions = np.genfromtxt('expert_data/cartpole/actions.csv',
dtype=np.int32)
expert_dataset = Dset_gym(inputs=expert_observations,
labels=expert_actions,
randomize=True)
# expert_dataset = (expert_observations, expert_actions)
reward_giver = Discriminator(env,
args.adversary_hidden_size,
entcoeff=args.adversary_entcoeff)
reward_guidance = Guidance(env,
args.policy_hidden_size,
expert_dataset=expert_dataset)
train(env, args.seed, policy_fn, reward_giver, reward_guidance,
expert_dataset, args.algo, args.g_step, args.d_step,
args.policy_entcoeff, args.num_timesteps, args.save_per_iter,
args.checkpoint_dir, args.log_dir, args.pretrained,
args.BC_max_iter, args.loss_percent, task_name)
elif args.task == 'evaluate':
avg_len, avg_ret = runner(env,
policy_fn,
args.load_model_path,
timesteps_per_batch=1024,
number_trajs=100,
stochastic_policy=args.stochastic_policy,
save=args.save_sample)
result = np.array([avg_ret, avg_len])
txt_name = args.load_model_path + 'result.txt'
np.savetxt(txt_name, result, fmt="%d", delimiter=" ")
print(args.load_model_path, avg_ret, avg_len)
print('保存成功')
else:
raise NotImplementedError
env.close()
def main(args):
U.make_session(num_cpu=1).__enter__()
set_global_seeds(args.seed)
env = gym.make(args.env_id)
task_name = get_task_short_name(args)
logger.configure(dir='log_trpo_mujoco/%s' % task_name)
def policy_fn(name, ob_space, ac_space, reuse=False):
return MlpPolicy(name=name,
ob_space=ob_space,
ac_space=ac_space,
reuse=reuse,
hid_size=args.policy_hidden_size,
num_hid_layers=2)
import logging
import os.path as osp
import bench
env = bench.Monitor(
env,
logger.get_dir() and osp.join(logger.get_dir(), "monitor.json"))
env.seed(args.seed)
gym.logger.setLevel(logging.WARN)
args.checkpoint_dir = osp.join(args.checkpoint_dir, task_name)
args.log_dir = osp.join(args.log_dir, task_name)
if args.task == 'train':
expert_dataset = Mujoco_Dset(expert_path=args.expert_path,
traj_limitation=args.traj_limitation)
reward_giver = Discriminator(env,
args.adversary_hidden_size,
entcoeff=args.adversary_entcoeff)
reward_guidance = Guidance(env,
args.policy_hidden_size,
expert_dataset=expert_dataset)
train(env, args.seed, policy_fn, reward_giver, reward_guidance,
expert_dataset, args.algo, args.g_step, args.d_step,
args.policy_entcoeff, args.num_timesteps, args.save_per_iter,
args.checkpoint_dir, args.log_dir, args.pretrained,
args.BC_max_iter, args.loss_percent, task_name)
elif args.task == 'evaluate':
avg_len, avg_ret = runner(env,
policy_fn,
args.load_model_path,
timesteps_per_batch=1024,
number_trajs=100,
stochastic_policy=args.stochastic_policy,
save=args.save_sample)
result = np.array([avg_ret, avg_len])
txt_name = args.load_model_path + 'result.txt'
np.savetxt(txt_name, result, fmt="%d", delimiter=" ")
print(args.load_model_path, avg_ret, avg_len)
print('保存成功')
else:
raise NotImplementedError
env.close()