def __init__(
self,
env,
n_bins_obs=10,
memory_size=100,
state_preprocess_fn=None,
state_preprocess_kwargs=None,
):
Wrapper.__init__(self, env)
if state_preprocess_fn is None:
assert isinstance(env.observation_space, spaces.Box)
assert isinstance(env.action_space, spaces.Discrete)
self.state_preprocess_fn = state_preprocess_fn or identity
self.state_preprocess_kwargs = state_preprocess_kwargs or {}
self.memory = TrajectoryMemory(memory_size)
self.total_visit_counter = DiscreteCounter(self.env.observation_space,
self.env.action_space,
n_bins_obs=n_bins_obs)
self.episode_visit_counter = DiscreteCounter(
self.env.observation_space,
self.env.action_space,
n_bins_obs=n_bins_obs)
self.current_state = None
self.curret_step = 0
def __init__(self, env):
Wrapper.__init__(self, env, wrap_spaces=True)
obs_space = self.env.observation_space
assert isinstance(obs_space, Discrete)
self.observation_space = Box(
low=0.0, high=1.0, shape=(obs_space.n,), dtype=np.uint32
)
def __init__(self, env, horizon):
"""
Parameters
----------
horizon: int
"""
Wrapper.__init__(self, env)
self.horizon = horizon
assert self.horizon >= 1
self.current_step = 0
def __init__(self,
env,
uncertainty_estimator_fn,
uncertainty_estimator_kwargs=None,
bonus_scale_factor=1.0,
bonus_max=np.inf):
Wrapper.__init__(self, env)
self.bonus_scale_factor = bonus_scale_factor
self.bonus_max = bonus_max
uncertainty_estimator_kwargs = uncertainty_estimator_kwargs or {}
uncertainty_estimator_fn = load(uncertainty_estimator_fn) if isinstance(uncertainty_estimator_fn, str) else \
uncertainty_estimator_fn
self.uncertainty_estimator = uncertainty_estimator_fn(
env.observation_space, env.action_space,
**uncertainty_estimator_kwargs)
self.previous_obs = None
def __init__(self, env):
Wrapper.__init__(self, env)
def __init__(self, env, reward_range):
Wrapper.__init__(self, env)
self.reward_range = reward_range
assert reward_range[0] < reward_range[1]
assert reward_range[0] > -np.inf and reward_range[1] < np.inf
