OakEmu-RL v0.1.0

Appearance

TrajectoryFrontierUtilityConfiguration ​

Cell score for the learning frontier: route_value × (1 − local_progress).

High where a win is reachable (terrain) but the policy cannot yet make the next hop (skill), so midpoint starts concentrate at the deepest cell the policy can't climb past and recede one hop at a time as local progress rises. Pairs a backpropagated route utility (read per bucket) with a local_progress utility, taking the route utility and the corroboration gate from it: a cell only scores once its local-progress outcomes reach the local-progress utility's support_threshold, so barely-measured cells cannot dominate selection.

Source

Properties ​

uses_cell_key_grouping ​

python
uses_cell_key_grouping: bool

Source

Values ​

kind ​

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

Source

local_progress_utility ​

python
local_progress_utility: str

Name of the local_progress utility supplying (1 − L), its route utility, and the gate.

Source