OakEmu-RL v0.1.0

Appearance

TrajectoryDiscoveryUtilityConfiguration ​

Cell score for discovery at a stuck boundary: traffic × (1 − local_progress).

High where the policy keeps arriving (high visit traffic) but cannot advance (low local progress) — that is the cells the from-start route piles up on and dies, like the post-item-5 drop. Unlike the frontier, it does not need a win to be reachable, so it can target never-crossed transitions the frontier cannot. Driving midpoint starts here sends the random-episode fraction to random-walk the missing transition and discover the first crossing; once a win crosses, the cell's win-proximity lights up and the frontier takes over consolidation.

traffic = visits / (visits + visit_half) saturates to 1 for heavily-visited cells, so it balances against (1 − L) ∈ [0, 1] instead of being dominated by raw visit counts.

Source

Properties ​

is_cell_scoring ​

python
is_cell_scoring: bool

Source

uses_cell_key_grouping ​

python
uses_cell_key_grouping: bool

Source

Values ​

kind ​

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

Source

local_progress_utility ​

python
local_progress_utility: str

Name of the local_progress utility supplying (1 − L) and its unknown_prior.

Source

visit_half ​

python
visit_half: float = Field(default=1000.0, gt=0.0)

Visit count at which the traffic factor reaches 0.5.

Source