Results about the output of a reward model #
Every state has finite absolute first and second output moments.
Equations
- model.IntegrableMoments = ∀ (i : Fin model.size), MeasureTheory.Integrable (fun (x : ℝ) => x) (model.outputAt i) ∧ MeasureTheory.Integrable (fun (x : ℝ) => x ^ 2) (model.outputAt i)
Instances For
def
Determinize.Spec.RewardModel.ResultMatches
(model : Model)
(program : Paper.Expr)
(statistics : FiniteModel.OutputStatistics)
:
An exported reward result includes integrability, not only integral equalities.
Equations
- Determinize.Spec.RewardModel.ResultMatches model program statistics = (model.Matches program ∧ statistics.Matches model.outputMeasure)