OakEmu-RL v0.1.0

Appearance

MaxCellValueAggregator ​

python
MaxCellValueAggregator(current: float = 0.0, initialized: bool = False) -> None

Persistent maximum over observed values.

A cell's downstream value should answer "how much win is reachable through here", which a running average drowns when wins are rare relative to visits: a cell visited tens of thousands of times with a 0.05% win rate averages to ~0 even though a winning route passes through it. Tracking the maximum keeps that cell lit so the receding-start curriculum can practise the cells one hop behind the mastered frontier; per-cell competence handles whether the policy has actually mastered them yet.

Source

Properties ​

value ​

python
value: float

Source

Methods ​

record ​

python
record(observation: float) -> None

Source

state_dict ​

python
state_dict() -> dict[str, Any]

Source

Values ​

current ​

python
current: float = 0.0

Source

initialized ​

python
initialized: bool = False

Source