ray.rllib.utils.exploration.epsilon_greedy.EpsilonGreedy.on_episode_start

EpsilonGreedy.on_episode_start(policy: Policy, *, environment: BaseEnv = None, episode: int = None, tf_sess: tf.Session | None = None)

Handles necessary exploration logic at the beginning of an episode.

Parameters:

• policy – The Policy object that holds this Exploration.

• environment – The environment object we are acting in.

• episode – The number of the episode that is starting.

• tf_sess – In case of tf, the session object.