def build_pend_env(args, **kwargs):
alg = args.alg
seed = args.seed
flatten_dict_observations = alg not in {'her'}
env = make_vec_env(args.env, 'classic_control', args.num_env or 1, seed, reward_scale=args.reward_scale,
flatten_dict_observations=flatten_dict_observations)
return VecVAEStack(env, k=k, load_from=vae_name.replace('/', ':'))
def build_pend_env(args, **kwargs):
alg = args.alg
seed = args.seed
flatten_dict_observations = alg not in {'her'}
env = make_vec_env(args.env, 'atari', args.num_env or 1, seed, reward_scale=args.reward_scale,
flatten_dict_observations=flatten_dict_observations)
return VecVAEStack(env, k=k, load_from=vae_name.replace('/', ':'), vae_network='atari', norm_fac=(1/255))
def build_pend_env(args, **kwargs):
alg = args.alg
seed = args.seed
flatten_dict_observations = alg not in {'her'}
env = make_vec_env(args.env,
'atari',
args.num_env or 1,
seed,
reward_scale=args.reward_scale,
flatten_dict_observations=flatten_dict_observations)
return VecVAEStack(env, k=3, load_from=vae_name)
def build_pend_env(args, **kwargs):
alg = args.alg
seed = args.seed
flatten_dict_observations = alg not in {'her'}
env = make_vec_env(args.env,
'classic_control',
args.num_env or 1,
seed,
reward_scale=args.reward_scale,
flatten_dict_observations=flatten_dict_observations)
return VecVAEStack(
env,
k=3,
load_from='pendvisualuniform-b77.5-lat5-lr0.001-2019-03-21T00/13'.
replace('/', ':'))
import numpy as np
from baselines.common.cmd_util import make_vec_env
from gym.envs.classic_control.pendulum_test import PendulumTestEnv
import matplotlib.pyplot as plt
import seaborn as sns
vae_name = 'pendvisualuniform-b77.5-lat5-lr0.001-2019-03-21T00/13'.replace(
'/', ':')
env = make_vec_env('PendulumTest-v0',
'classic_control',
2,
0,
flatten_dict_observations=False)
venv = VecVAEStack(env, k=3, load_from=vae_name)
thetas = np.linspace(0, 2 * np.pi, 20)
dings1 = []
dings2 = []
o = venv.reset()
d = False
while not np.any(d):
print(o)
exit(0)
# when using env obs, we can plot to sanity check theta
# plt.imshow(np.squeeze(o), cmap='Greys_r', label='theta {}'.format(th))
# plt.title('{}'.format(th))