# We are done as per our max-episode-len.
if self.max_episode_len is not None and \
self.steps >= self.max_episode_len:
done = True
# Max not reached yet -> Sample done via p_done.
elif self.p_done > 0.0:
done = bool(
np.random.choice([True, False],
p=[self.p_done, 1.0 - self.p_done]))
return self.observation_space.sample(), \
float(self.reward_space.sample()), done, {}
# Multi-agent version of the RandomEnv.
RandomMultiAgentEnv = make_multi_agent(lambda c: RandomEnv(c))
# Large observation space "pre-compiled" random env (for testing).
class RandomLargeObsSpaceEnv(RandomEnv):
def __init__(self, config=None):
config = config or {}
config.update(
{"observation_space": gym.spaces.Box(-1.0, 1.0, (5000, ))})
super().__init__(config=config)
# Large observation space + cont. actions "pre-compiled" random env
# (for testing).
class RandomLargeObsSpaceEnvContActions(RandomEnv):
def __init__(self, config=None):
@override(CartPoleEnv)
def reset(self):
# Reset timestep counter for the new episode.
self.timesteps = 0
# Should we crash?
if np.random.random() < self.p_crash_reset or (
self.crash_after_n_steps is not None
and self.crash_after_n_steps == 0):
raise EnvError(
"Simulated env crash in `reset()`! Feel free to use any "
"other exception type here instead.")
return super().reset()
@override(CartPoleEnv)
def step(self, action):
# Increase timestep counter for the ongoing episode.
self.timesteps += 1
# Should we crash?
if np.random.random() < self.p_crash or (self.crash_after_n_steps
and self.crash_after_n_steps
== self.timesteps):
raise EnvError(
"Simulated env crash in `step()`! Feel free to use any "
"other exception type here instead.")
# No crash.
return super().step(action)
MultiAgentCartPoleCrashing = make_multi_agent(
lambda config: CartPoleCrashing(config))