Language

unilab.base.curriculum¶

Curriculum learning for adaptive difficulty adjustment.

Classes

`EpisodeLengthTracker`	Track moving average of episode length.
`PenaltyCurriculum`	Adaptive penalty scaling based on episode length.

class unilab.base.curriculum.EpisodeLengthTracker[source]¶

Bases: object

Track moving average of episode length.

Parameters:

num_envs (int)
window_size (int)

__init__(num_envs, window_size=1000)[source]¶

Parameters:

num_envs (int)
window_size (int)

update(episode_lengths)[source]¶

Update average with new episode lengths.

Parameters:: episode_lengths (ndarray)
Return type:: None

class unilab.base.curriculum.PenaltyCurriculum[source]¶

Bases: object

Adaptive penalty scaling based on episode length.

Parameters:

env (Any)
enabled (bool)
initial_scale (float)
min_scale (float)
max_scale (float)
level_down_threshold (float)
level_up_threshold (float)
degree (float)

__init__(env, enabled=True, initial_scale=0.5, min_scale=0.5, max_scale=1.0, level_down_threshold=150.0, level_up_threshold=750.0, degree=0.001)[source]¶

Parameters:

env (Any)
enabled (bool)
initial_scale (float)
min_scale (float)
max_scale (float)
level_down_threshold (float)
level_up_threshold (float)
degree (float)

update(average_episode_length)[source]¶

Update penalty scale based on average episode length.

Parameters:: average_episode_length (float)
Return type:: None