def main():
args = control_arg_parser().parse_args()
if args.env == 'LunarLanderContinuousPOMDP-v0':
newenv(hist_len=args.hist_len)
if args.train is True:
ENV_path = get_dir(os.path.join(args.log_dir, args.env))
log_dir = os.path.join(ENV_path, args.method +"-"+
'{0}'.format(args.seed))+"-" +\
datetime.datetime.now().strftime("%m-%d-%H-%M")
# if MPI.COMM_WORLD.Get_rank() == 0:
logger.configure(dir=log_dir)
save_args(args)
train(args.env,
num_timesteps=args.num_timesteps,
seed=args.seed,
nsteps=args.nsteps,
batch_size=args.batch_size,
epoch=args.epoch,
method=args.method,
net_size=args.net_size,
ncpu=args.ncpu,
i_trial=args.seed,
load_path=args.load_path,
use_entr=int(args.use_entr))
if args.render is True:
video_path = osp.split(osp.split(args.load_path)[0])[0]
render(args.env,
nsteps=args.nsteps,
batch_size=args.batch_size,
net_size=args.net_size,
load_path=args.load_path,
video_path=video_path,
iters=args.iters)
def main():
args = control_arg_parser().parse_args()
# rank = MPI.COMM_WORLD.Get_rank()
# print("This is rank:", rank, "Seed:", args.seed)
if args.env == 'LunarLanderContinuousPOMDP-v0':
newenv(hist_len=args.hist_len,
block_high=float(args.block_high),
policy_name=args.policy_name)
# if args.train is True:
ENV_path = get_dir(os.path.join(args.log_dir, args.env))
log_dir = os.path.join(ENV_path, args.method +"-"+
'{}'.format(args.seed))+"-" +\
datetime.datetime.now().strftime("%m-%d-%H-%M")
logger.configure(dir=log_dir)
# logger.log("This is rank {}".format(rank))
save_args(args)
train(args.env,
num_timesteps=args.num_timesteps,
seed=args.seed,
hist_len=args.hist_len,
nsteps=args.nsteps,
batch_size=args.batch_size,
epoch=args.epoch,
env_name=args.env,
method=args.method,
net_size=tuple(args.net_size),
ncpu=args.ncpu,
policy_name=args.policy_name,
load_path=args.load_path,
use_entr=int(args.use_entr),
rank=args.seed,
checkpoint=args.checkpoint,
filter_size=args.filter_size)
def make_control_env(env_id, seed, hist_len, block_high, version0, give_state):
"""
Create a wrapped, monitored gym.Env for MuJoCo.
"""
set_global_seeds(seed)
if env_id == 'LunarLanderContinuousPOMDP-v0':
newenv(hist_len=hist_len,
block_high=block_high,
version0=version0,
give_state=give_state)
env = gym.make(env_id)
env = Monitor(env,
logger.get_dir(),
allow_early_resets=True,
version0=version0)
env.seed(seed)
return env
# from baselines.env.box2d.lunar_lander_pomdp import LunarLanderContinuousPOMDP
from baselines import bench, logger
from baselines.env.envsetting import newenv
import gym
newenv(hist_len=5, block_high=5/8)
env = gym.make('LunarLanderContinuousPOMDP-v0')
# env = LunarLanderContinuousPOMDP(hist_len=0)
env = bench.Monitor(env, logger.get_dir())
obs,state = env.reset()
ob_shapce = env.observation_space
total_shape = env.total_space
print("obs:{0}, state:{1}".format(obs, state))
print("total space:", total_shape, total_shape.shape)
print("obs space:", ob_shapce, ob_shapce.shape)
l = 0
# while True:
# env.render(mode="rgb_array")
# ac = env.action_space.sample()
# [obs,state], rwd, done, _ = env.step(ac)
# print("obs:{0}, state:{1}".format(obs, state))
# if done:
# break
# l+=1
# print("Episode Length from baselines:{}".format(l))
import pandas as pd
import imageio
import numpy as np
import joblib
import tensorflow as tf
from baselines import bench, logger
from baselines.common.vec_env.vec_normalize import VecNormalize
from baselines.common.vec_env.dummy_vec_env import DummyVecEnv
from baselines.env.envsetting import newenv
save_path = '/home/zhi/Documents/ReinforcementLearning/tmp'
hist_len = 10
block_hight = 5 / 8
newenv(hist_len=hist_len, block_high=block_hight, policy_name='MlpPolicy')
def make_env():
env = gym.make('LunarLanderContinuousPOMDP-v0')
env = bench.Monitor(env,
os.path.join(save_path, 'render-result'),
allow_early_resets=True)
return env
env = DummyVecEnv([make_env])
env = VecNormalize(env)
ob = env.reset()