def test_gray_scale_observation(env_id, keep_dim):
gray_env = AtariPreprocessing(gym.make(env_id),
screen_size=84,
grayscale_obs=True)
rgb_env = AtariPreprocessing(gym.make(env_id),
screen_size=84,
grayscale_obs=False)
wrapped_env = GrayScaleObservation(rgb_env, keep_dim=keep_dim)
assert rgb_env.observation_space.shape[-1] == 3
seed = 0
gray_env.seed(seed)
wrapped_env.seed(seed)
gray_obs = gray_env.reset()
wrapped_obs = wrapped_env.reset()
if keep_dim:
assert wrapped_env.observation_space.shape[-1] == 1
assert len(wrapped_obs.shape) == 3
wrapped_obs = wrapped_obs.squeeze(-1)
else:
assert len(wrapped_env.observation_space.shape) == 2
assert len(wrapped_obs.shape) == 2
# ALE gray scale is slightly different, but no more than by one shade
assert np.allclose(gray_obs.astype("int32"),
wrapped_obs.astype("int32"),
atol=1)
if environment_used != "CartPole-v1":
env = GrayScaleObservation(env)
env = ResizeObservation(env, resolution)
env = FrameStack(env, 4)
agent = VizdoomAgent(env.action_space, resolution , eta, test_mode, environment_used)
else:
agent = RandomAgent(env.action_space, env.observation_space, eta, test_mode, environment_used)
# You provide the directory to write to (can be an existing
# directory, including one with existing data -- all monitor files
# will be namespaced). You can also dump to a tempdir if you'd
# like: tempfile.mkdtemp().
outdir = '/tmp/random-agent-results'
env = wrappers.Monitor(env, directory=outdir, force=True)
env.seed(0)
mem = Memory()
episode_count = 50
reward = 0
done = False
rewardPerEp = numpy.array([])
ep = numpy.array([])
interaction = numpy.array([])
for i in range(episode_count):
env.reset()
actual_state, reward, done, info = env.step(env.action_space.sample())