def test_normalize_action():
env = gym.make('CartPole-v1')
with pytest.raises(AssertionError):
env = NormalizeAction(env)
del env
env = gym.make('Pendulum-v0')
env = NormalizeAction(env)
env.reset()
with pytest.raises(AssertionError):
env.step(10 + env.action_space.sample())
def make_env(config, seed, mode):
assert mode in ['train', 'eval']
env = gym.make(config['env.id'])
env.seed(seed)
env.observation_space.seed(seed)
env.action_space.seed(seed)
env = NormalizeAction(
env) # TODO: use gym new wrapper RescaleAction when it's merged
if mode == 'eval':
env = RecordEpisodeStatistics(env, deque_size=100)
env = TimeStepEnv(env)
return env
def _make_env():
env = gym.make(config['env.id'])
env = env.env # strip out gym TimeLimit, TODO: remove until gym update it
env = TimeLimit(env, env.spec.max_episode_steps)
env = NormalizeAction(env)
return env