ray.rllib.utils.exploration.ornstein_uhlenbeck_noise.OrnsteinUhlenbeckNoise.__init__#

OrnsteinUhlenbeckNoise.__init__(action_space, *, framework: str, ou_theta: float = 0.15, ou_sigma: float = 0.2, ou_base_scale: float = 0.1, random_timesteps: int = 1000, initial_scale: float = 1.0, final_scale: float = 0.02, scale_timesteps: int = 10000, scale_schedule: Schedule | None = None, **kwargs)[source]#

Initializes an Ornstein-Uhlenbeck Exploration object.

Parameters:

  • action_space – The gym action space used by the environment.

  • ou_theta – The theta parameter of the Ornstein-Uhlenbeck process.

  • ou_sigma – The sigma parameter of the Ornstein-Uhlenbeck process.

  • ou_base_scale – A fixed scaling factor, by which all OU- noise is multiplied. NOTE: This is on top of the parent GaussianNoise’s scaling.

  • random_timesteps – The number of timesteps for which to act completely randomly. Only after this number of timesteps, the self.scale annealing process will start (see below).

  • initial_scale – The initial scaling weight to multiply the noise with.

  • final_scale – The final scaling weight to multiply the noise with.

  • scale_timesteps – The timesteps over which to linearly anneal the scaling factor (after(!) having used random actions for random_timesteps steps.

  • scale_schedule – An optional Schedule object to use (instead of constructing one from the given parameters).

  • framework – One of None, “tf”, “torch”.