def setup(exp, single_threaded):
import lab
lab.undo_logger_setup()
from . import policies, tf_util
config = Config(**exp['config'])
env = lab.make(exp['env_id'])
sess = make_session(single_threaded=single_threaded)
policy = getattr(policies, exp['policy']['type'])(env.observation_space,
env.action_space,
**exp['policy']['args'])
tf_util.initialize()
return config, env, sess, policy
def test_smoke(env_id):
"""Check that environments start up without errors and that we can extract rewards and observations"""
lab.undo_logger_setup()
logging.getLogger().setLevel(logging.INFO)
env = lab.make(env_id)
if env.metadata.get('configure.required', False):
if os.environ.get('FORCE_LATEST_INSTITUTE_DOCKER_RUNTIMES'): # Used to test institute-envs in CI
configure_with_latest_docker_runtime_tag(env)
else:
env.configure(remotes=1)
env = wrappers.Unvectorize(env)
env.reset()
_rollout(env, timestep_limit=60*30) # Check a rollout
def test_nice_vnc_semantics_match(spec, matcher, wrapper):
# Check that when running over VNC or using the raw environment,
# semantics match exactly.
lab.undo_logger_setup()
logging.getLogger().setLevel(logging.INFO)
spaces.seed(0)
vnc_env = spec.make()
if vnc_env.metadata.get('configure.required', False):
vnc_env.configure(remotes=1)
vnc_env = wrapper(vnc_env)
vnc_env = wrappers.Unvectorize(vnc_env)
env = lab.make(spec._kwargs['lab_core_id'])
env.seed(0)
vnc_env.seed(0)
# Check that reset observations work
reset(matcher, env, vnc_env, stage='initial reset')
# Check a full rollout
rollout(matcher, env, vnc_env, timestep_limit=50, stage='50 steps')
# Reset to start a new episode
reset(matcher, env, vnc_env, stage='reset to new episode')
# Check that a step into the next episode works
rollout(matcher,
env,
vnc_env,
timestep_limit=1,
stage='1 step in new episode')
# Make sure env can be reseeded
env.seed(1)
vnc_env.seed(1)
reset(matcher, env, vnc_env, 'reseeded reset')
rollout(matcher, env, vnc_env, timestep_limit=1, stage='reseeded step')
import numpy as np
import lab
from lab import spaces, envs
lab.undo_logger_setup()
import logging
logging.getLogger('lab.core').addHandler(logging.NullHandler())
num_trials = 50
print 'Name & Random policy performance'
names = [
'CartPole-v0', 'Acrobot-v0', 'MountainCar-v0', 'Reacher-v1',
'HalfCheetah-v1', 'Hopper-v1', 'Walker2d-v1', 'Ant-v1', 'Humanoid-v1'
]
for env_name in names:
env = envs.make(env_name)
returns = []
for _ in xrange(num_trials):
env.reset()
ret = 0.
for _ in xrange(env.spec.timestep_limit):
_, r, done, _ = env.step(env.action_space.sample())
ret += r
if done: break
returns.append(ret)
print '{} & {} \pm {}'.format(env_name, np.mean(returns), np.std(returns))