def test_pickleable(self):
env = IdentificationEnv(CartpoleEnv, dict(obs_noise=5.))
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert env.mdp_args == round_trip.mdp_args
assert round_trip.env.obs_noise == env.env.obs_noise
step_env(round_trip)
def test_all_mt50():
mt50 = MT50()
train_env_instances = {
env_name: env_cls()
for (env_name, env_cls) in mt50.train_classes.items()
}
train_env_rand_vecs = check_tasks_unique(mt50.train_tasks,
mt50._train_classes.keys())
for task in mt50.train_tasks:
env = train_env_instances[task.env_name]
env.set_task(task)
obs = env.reset()
assert np.any(obs[-3:] != np.array([0, 0, 0]))
assert env.observation_space.shape == (39, )
assert env.random_init == True
old_obj_init = env.obj_init_pos
old_target_pos = env._target_pos
step_env(env, max_path_length=STEPS, render=False)
assert np.all(np.allclose(old_obj_init, env.obj_init_pos))
assert np.all(np.allclose(old_target_pos, env._target_pos))
# only needs to be done for 50 environments once
check_target_poss_unique(train_env_instances, train_env_rand_vecs)
for env in train_env_instances.values():
env.close()
del train_env_instances
assert len(mt50.test_classes) == 0
assert len(mt50.test_tasks) == 0
train_test_rand_vecs = set()
for rand_vecs in train_env_rand_vecs.values():
for rand_vec in rand_vecs:
train_test_rand_vecs.add(tuple(rand_vec))
assert len(train_test_rand_vecs) == 50 * 50
def test_pickleable(self):
env = Swimmer3DHillEnv(regen_terrain=False)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.difficulty == env.difficulty
step_env(round_trip)
def test_can_step(env_ids):
"""Test Bullet environments can step"""
for env_id in env_ids:
# extract id string
env_id = env_id.replace('- ', '')
if env_id == 'KukaCamBulletEnv-v0':
# Kuka environments calls py_bullet.resetSimulation() in reset()
# unconditionally, which globally resets other simulations. So
# only one Kuka environment is tested.
continue
if env_id in _get_unsupported_env_list():
pytest.skip('Skip unsupported Bullet environments')
env = BulletEnv(env_name=env_id)
ob_space = env.observation_space
act_space = env.action_space
env.reset()
ob = ob_space.sample()
assert ob_space.contains(ob)
a = act_space.sample()
assert act_space.contains(a)
# Skip rendering because it causes TravisCI to run out of memory
step_env(env, render=False)
env.close()
def test_hidden_goal_envs():
for env_key, env_cls in ALL_V2_ENVIRONMENTS_GOAL_HIDDEN.items():
assert "goal-hidden" in env_key
assert "GoalHidden" in env_cls.__name__
state_before = np.random.get_state()
env = env_cls(seed=5)
env2 = env_cls(seed=5)
step_env(env, max_path_length=3, iterations=3, render=False)
first_target = env._target_pos
env.reset()
second_target = env._target_pos
assert (first_target == second_target).all()
env.reset()
env2.reset()
assert (env._target_pos == env2._target_pos).all()
state_after = np.random.get_state()
for idx, (state_before_idx,
state_after_idx) in enumerate(zip(state_before,
state_after)):
if idx == 1:
assert (state_before_idx == state_after_idx).all()
else:
assert state_before_idx == state_after_idx
def test_pickleable(self):
env = PointEnv()
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
step_env(round_trip)
env.close()
round_trip.close()
def test_all_mt10():
mt10 = MT10()
train_env_instances = {
env_name: env_cls()
for (env_name, env_cls) in mt10.train_classes.items()
}
train_env_rand_vecs = check_tasks_unique(mt10.train_tasks,
mt10._train_classes.keys())
for task in mt10.train_tasks:
env = train_env_instances[task.env_name]
env.set_task(task)
env.reset()
old_obj_init = env.obj_init_pos
old_target_pos = env._target_pos
step_env(env, max_path_length=STEPS, render=False)
assert np.all(np.allclose(old_obj_init, env.obj_init_pos))
assert np.all(np.allclose(old_target_pos, env._target_pos))
for env in train_env_instances.values():
env.close()
del train_env_instances
assert len(mt10.test_classes) == 0
assert len(mt10.test_tasks) == 0
train_test_rand_vecs = set()
for rand_vecs in train_env_rand_vecs.values():
for rand_vec in rand_vecs:
train_test_rand_vecs.add(tuple(rand_vec))
assert len(train_test_rand_vecs) == 10 * 50
def test_pickleable(self):
env = AntGatherEnv(n_apples=1)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.n_apples == env.n_apples
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
env = SwimmerEnv(ctrl_cost_coeff=1.0)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.ctrl_cost_coeff == env.ctrl_cost_coeff
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
env = CarParkingEnv(random_start_range=2.0)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.random_start_range == env.random_start_range
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
env = GridWorldEnv(desc="8x8")
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.start_state == env.start_state
step_env(round_trip)
round_trip.close()
env.close()
def test_all_pickleable(self, domain_name, task_name):
env = DmControlEnv.from_suite(domain_name, task_name)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
# Skip rendering because it causes TravisCI to run out of memory
step_env(round_trip, render=False)
round_trip.close()
env.close()
def test_can_step_and_render(self):
ob_space = self.env.observation_space
act_space = self.env.action_space
ob = self.env.reset()
assert ob_space.contains(ob)
a = act_space.sample()
assert act_space.contains(a)
step_env(self.env, n=10, render=True)
def test_pickleable(self):
env = Walker2DEnv(ctrl_cost_coeff=3.)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.ctrl_cost_coeff == env.ctrl_cost_coeff
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
inner_env = CartpoleEnv(obs_noise=5.)
env = SlidingMemEnv(inner_env, n_steps=10)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.n_steps == env.n_steps
assert round_trip.env.obs_noise == env.env.obs_noise
step_env(round_trip)
def test_pickleable(self):
env = CartpoleSwingupEnv(obs_noise=1.0)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.obs_noise == env.obs_noise
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
env = SimpleHumanoidEnv(alive_bonus=1.0)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.alive_bonus == env.alive_bonus
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
inner_env = CartpoleEnv(frame_skip=10)
env = NoisyObservationEnv(inner_env, obs_noise=5.)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.obs_noise == env.obs_noise
assert round_trip.env.frame_skip == env.env.frame_skip
step_env(round_trip)
def test_pickleable(self):
env = SwimmerMazeEnv(n_bins=2)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip._n_bins == env._n_bins
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
env = HalfCheetahHillEnv(regen_terrain=False)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.difficulty == env.difficulty
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
inner_env = CartpoleEnv(frame_skip=10)
env = DelayedActionEnv(inner_env, action_delay=10)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.action_delay == env.action_delay
assert round_trip.env.frame_skip == env.env.frame_skip
step_env(round_trip)
def test_pickleable(self):
inner_env = CartpoleEnv(obs_noise=5.)
env = NormalizedEnv(inner_env, scale_reward=10.)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip._scale_reward == env._scale_reward
assert round_trip.env.obs_noise == env.env.obs_noise
step_env(round_trip)
def test_pickleable(self):
env = AntEnv(action_noise=1.0)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.action_noise == env.action_noise
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
env = HopperEnv(alive_coeff=2.0)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip.alive_coeff == env.alive_coeff
step_env(round_trip)
round_trip.close()
env.close()
def test_pickleable(self):
inner_env = CartpoleEnv(obs_noise=5.)
env = OcclusionEnv(inner_env, [1])
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
obs = inner_env.reset()
assert round_trip.occlude(obs) == env.occlude(obs)
assert round_trip.env.obs_noise == env.env.obs_noise
step_env(round_trip)
def test_can_step_and_render(self, domain_name, task_name):
env = DmControlEnv(domain_name, task_name)
ob_space = env.observation_space
act_space = env.action_space
ob = env.reset()
assert ob_space.contains(ob)
a = act_space.sample()
assert act_space.contains(a)
step_env(env, n=10, render=True)
def test_all_mt50():
mt50_env = MT50()
tasks = mt50_env.sample_tasks(51)
for t in tasks:
mt50_env.set_task(t)
step_env(mt50_env, max_path_length=3)
mt50_env.close()
del mt50_env
def test_pickleable(self):
env = IdentificationEnv(PointEnv, dict(goal=(1., 2.)))
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert env.mdp_args == round_trip.mdp_args
assert np.array_equal(round_trip.env._goal, env.env._goal)
step_env(round_trip)
round_trip.close()
env.close()
def test_all_pickleable(self, domain_name, task_name):
env = DMControlEnv.from_suite(domain_name, task_name)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
# Skip rendering because it causes TravisCI to run out of memory
# Sometimes random actions lead to physics errors
with env._env.physics.suppress_physics_errors():
step_env(env, visualize=False)
round_trip.close()
env.close()
def test_pickleable(self):
inner_env = PointEnv(goal=(1., 2.))
env = NormalizedEnv(inner_env, scale_reward=10.)
round_trip = pickle.loads(pickle.dumps(env))
assert round_trip
assert round_trip._scale_reward == env._scale_reward
assert np.array_equal(round_trip.env._goal, env.env._goal)
step_env(round_trip)
round_trip.close()
env.close()
