OakEmu-RL v0.1.0

Appearance

TrajectoryCompetenceUtilityConfiguration ​

Utility that scores cells by how learnable they currently are.

Tracks an EMA of episode success per start cell and weights cells by success_rate^success_power * (1 - success_rate)^failure_power, which peaks where the policy partially succeeds: the learning frontier. Mastered cells and hopeless cells both fade out, so start selection naturally recedes along a winning route as mastery spreads.

Source

Properties ​

is_cell_scoring ​

python
is_cell_scoring: bool

Source

uses_cell_key_grouping ​

python
uses_cell_key_grouping: bool

Source

wants_competence_table ​

python
wants_competence_table: bool

Source

Methods ​

competence_table_aggregator ​

python
competence_table_aggregator() -> TrajectoryCellValueAggregatorConfiguration | None

Source

Values ​

ema_alpha ​

python
ema_alpha: float = Field(default=0.05, gt=0.0, le=1.0)

Source

failure_power ​

python
failure_power: float = Field(default=1.0, ge=0.0)

Source

kind ​

python
kind: Literal['competence'] = 'competence'

Source

success ​

python
success: TrajectoryUtilityMetricConfiguration

Source

success_minimum ​

python
success_minimum: float = 0.5

Source

success_power ​

python
success_power: float = Field(default=1.0, ge=0.0)

Source

unknown_prior ​

python
unknown_prior: float = Field(default=0.5, ge=0.0, le=1.0)

Source