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_does_not_modify_action(self):
env = GridWorldEnv(desc='8x8')
a = env.action_space.sample()
a_copy = a
env.reset()
env.step(a)
assert a == a_copy
env.close()
def test_does_not_modify_action(self):
env = GridWorldEnv(desc="8x8")
a = env.action_space.sample()
a_copy = a
env.reset()
env.step(a)
self.assertEqual(a, a_copy)
env.close()
def test_obtain_samples(ray_local_session_fixture):
del ray_local_session_fixture
env = GridWorldEnv(desc='4x4')
policy = ScriptedPolicy(
scripted_actions=[2, 2, 1, 0, 3, 1, 1, 1, 2, 2, 1, 1, 1, 2, 2, 1])
algo = Mock(env_spec=env.spec, policy=policy, max_episode_length=16)
assert ray.is_initialized()
workers = WorkerFactory(seed=100,
max_episode_length=algo.max_episode_length,
n_workers=8)
sampler1 = RaySampler.from_worker_factory(workers, policy, env)
sampler2 = LocalSampler.from_worker_factory(workers, policy, env)
eps1 = sampler1.obtain_samples(0, 1000,
tuple(algo.policy.get_param_values()))
eps2 = sampler2.obtain_samples(0, 1000,
tuple(algo.policy.get_param_values()))
assert eps1.observations.shape[0] >= 1000
assert eps1.actions.shape[0] >= 1000
assert (sum(eps1.rewards[:eps1.lengths[0]]) == sum(
eps2.rewards[:eps2.lengths[0]]) == 1)
true_obs = np.array([0, 1, 2, 6, 10, 14])
true_actions = np.array([2, 2, 1, 1, 1, 2])
true_rewards = np.array([0, 0, 0, 0, 0, 1])
start = 0
for length in eps1.lengths:
observations = eps1.observations[start:start + length]
actions = eps1.actions[start:start + length]
rewards = eps1.rewards[start:start + length]
assert np.array_equal(observations, true_obs)
assert np.array_equal(actions, true_actions)
assert np.array_equal(rewards, true_rewards)
start += length
sampler1.shutdown_worker()
sampler2.shutdown_worker()
env.close()