Skill¶

Parameters¶

class agilerl.wrappers.learning.Skill(env: Env)¶

The Skill class, used in curriculum learning to teach agents skills. This class works as a wrapper around an environment that alters the reward to encourage learning of a particular skill.

Parameters:: env (Gymnasium-style environment) – Environment to learn in

skill_reward(observation: Any, reward: float, terminated: bool, truncated: bool, info: dict[str, Any]) → tuple[Any, float, bool, bool, dict[str, Any]]¶

Calculate the reward for the given observation, reward, terminated, truncated, and info.

Parameters:

observation (Any) – Observation
reward (float) – Reward
terminated (bool) – Terminated
truncated (bool) – Truncated
info (dict[str, Any]) – Info

Returns:

Tuple of (observation, reward, terminated, truncated, info)

Return type:

tuple[Any, float, bool, bool, dict[str, Any]]

step(action: Any) → tuple[Any, float, bool, bool, dict[str, Any]]¶

Step the environment and return the observation, reward, terminated, truncated, and info.

Parameters:: action (Any) – Action
Returns:: Tuple of (observation, reward, terminated, truncated, info)
Return type:: tuple[Any, float, bool, bool, dict[str, Any]]

BanditEnv¶

Parameters¶

class agilerl.wrappers.learning.BanditEnv(features: DataFrame, targets: DataFrame)¶

The Bandit learning environment class. Turns a labelled dataset into a reinforcement learning, Gym-style environment.

Parameters:

features (pd.DataFrame) – Dataset features
targets (pd.DataFrame) – Dataset targets corresponding to features

reset() → ndarray¶

Reset the environment and return the initial state.

Returns:: Initial state
Return type:: np.ndarray

step(k: int) → tuple[ndarray, float]¶

Step the environment and return the state and reward.

Parameters:: k (int) – Action
Returns:: Tuple of (state, reward)
Return type:: tuple[np.ndarray, float]