def test_flatten_observation(self):
gym_env = gym.make('Pong-v0')
env = GymWrapper(gym_env)
obs = env.reset()
assert obs.shape == (210, 160, 3)
env = FlattenObservation(env)
obs = env.reset()
assert obs.shape == (210 * 160 * 3, )
def test_scale_image_observation(self):
env = gym.make('Pong-v0')
env = GymWrapper(env)
env = ScaleImageObservation(env)
assert np.allclose(env.observation_space.high, 1.0)
assert np.allclose(env.observation_space.low, 0.0)
obs = env.reset()
assert np.alltrue(obs <= 1.0) and np.alltrue(obs >= 0.0)
obs, _, _, _ = env.step(env.action_space.sample())
assert np.alltrue(obs <= 1.0) and np.alltrue(obs >= 0.0)
def test_clip_reward(self, env_id):
env = gym.make(env_id)
env = GymWrapper(env)
env = ClipReward(env)
env.reset()
for _ in range(100):
obs, reward, done, info = env.step(env.action_space.sample())
assert reward >= -1.0 and reward <= 1.0
if done:
break
def test_frame_stack(self):
env = gym.make('CartPole-v1')
env = GymWrapper(env)
env = FrameStack(env, num_stack=4)
env.seed(1)
assert isinstance(env, Env) and isinstance(env, FrameStack)
assert env.num_stack == 4
assert env.observation_space.shape == (4, 4)
assert isinstance(env.stack_buffer, np.ndarray)
assert env.stack_buffer.shape == (4, 4)
assert np.all(env.stack_buffer == 0.0)
assert env.stack_buffer.dtype == np.float32
assert env.reset().shape == (4, 4)
obs = env.step(0)[0]
assert obs[:, 0].sum() != 0.0
assert obs[:, 1].sum() != 0.0
assert np.all(obs[:, 2:] == 0.0)
assert np.any(obs[:, 0] != obs[:, 1])
obs = env.step(1)[0]
obs = env.step(1)[0]
assert np.allclose(obs[:, -1],
[0.03073904, 0.00145001, -0.03088818, -0.03131252])
assert np.allclose(obs[:, 2],
[0.03076804, -0.19321568, -0.03151444, 0.25146705])
obs = env.step(1)[0]
assert np.allclose(obs[:, -1],
[0.03076804, -0.19321568, -0.03151444, 0.25146705])
def test_env_spec(self):
env = gym.make('CartPole-v1')
env = GymWrapper(env)
env.seed(0)
env_spec = EnvSpec(env)
assert isinstance(env_spec.observation_space, Box)
assert isinstance(env_spec.action_space, Discrete)
assert env_spec.control_type == 'Discrete'
assert env_spec.T == 500
assert env_spec.max_episode_reward == 475.0
assert env_spec.reward_range == (-float('inf'), float('inf'))
def test_reward_scale(self):
env = gym.make('CartPole-v1')
env = GymWrapper(env)
env = RewardScale(env, scale=0.02)
env.seed(1)
env.reset()
observation, reward, done, info = env.step(env.action_space.sample())
assert reward == 0.02
def test_time_aware_observation(self, env_id):
gym_env = gym.make(env_id)
env = GymWrapper(gym_env)
timed_env = TimeAwareObservation(env)
assert env.observation_space.shape[
0] + 1 == timed_env.observation_space.shape[0]
obs = env.reset()
timed_obs = timed_env.reset()
assert obs.shape[0] + 1 == timed_obs.shape[0]
assert timed_obs[-1] == 0.0
assert timed_env.t == 0.0
timed_obs, _, _, _ = timed_env.step(timed_env.action_space.sample())
assert obs.shape[0] + 1 == timed_obs.shape[0]
assert timed_obs[-1] == 1.0
assert timed_env.t == 1.0
timed_obs, _, _, _ = timed_env.step(timed_env.action_space.sample())
assert timed_obs[-1] == 2.0
assert timed_env.t == 2.0
timed_obs = timed_env.reset()
assert timed_obs[-1] == 0.0
assert timed_env.t == 0.0
def test_gym_wrapper(self):
gym_env = gym.make('CartPole-v1')
env = GymWrapper(gym_env)
assert isinstance(env, GymWrapper) and isinstance(env, Wrapper)
assert isinstance(gym_env, gym.Env) and isinstance(
env.unwrapped, gym.Env)
assert isinstance(env.env, gym.Wrapper)
assert env.reset().shape == (4, )
assert isinstance(env.observation_space, Box)
assert isinstance(env.action_space, Discrete)
assert len(env.step(env.action_space.sample())) == 4
assert env.seed(3) == [3]
assert env.T == 500
assert env.max_episode_reward == 475.0
assert env.reward_range == (-float('inf'), float('inf'))
del gym_env
del env
gym_env = gym.make('Pendulum-v0')
env = GymWrapper(gym_env)
assert isinstance(env, GymWrapper) and isinstance(env, Wrapper)
assert isinstance(gym_env, gym.Env) and isinstance(
env.unwrapped, gym.Env)
assert isinstance(env.env, gym.Wrapper)
assert env.reset().shape == (3, )
assert isinstance(env.observation_space, Box)
assert isinstance(env.action_space, Box)
assert len(env.step(env.action_space.sample())) == 4
assert env.seed(3) == [3]
assert env.T == 200
assert env.max_episode_reward is None
assert env.reward_range == (-float('inf'), float('inf'))
del gym_env
del env