noncomputable def
Determinize.Proof.RewardModel.momentAt
(model : Spec.RewardModel.Model)
(moment : FiniteModel.Moment)
(i : Fin model.size)
:
Equations
Instances For
theorem
Determinize.Proof.RewardModel.momentAt_equation
(model : Spec.RewardModel.Model)
(moment : FiniteModel.Moment)
(i : Fin model.size)
:
momentAt model moment i = match model.kind i with
| Spec.FiniteModel.StateKind.returned b => moment.real ↑b
| Spec.FiniteModel.StateKind.rejected => 0
| Spec.FiniteModel.StateKind.transient =>
(List.map
(fun (e : Spec.RewardModel.Edge model.size) =>
↑e.probability * match moment with
| FiniteModel.Moment.mass => momentAt model FiniteModel.Moment.mass e.target
| FiniteModel.Moment.first =>
momentAt model FiniteModel.Moment.first e.target + ↑e.reward * momentAt model FiniteModel.Moment.mass e.target
| FiniteModel.Moment.second =>
momentAt model FiniteModel.Moment.second e.target + 2 * ↑e.reward * momentAt model FiniteModel.Moment.first e.target + ↑e.reward ^ 2 * momentAt model FiniteModel.Moment.mass e.target)
(model.edges i)).sum
theorem
Determinize.Proof.RewardModel.eq_of_linear_equations
(model : Spec.FiniteModel.Model)
(paths : FiniteModel.Paths model)
(valid : FiniteModel.Paths.Valid model paths)
(rhs v w : Fin model.size → ℝ)
(hv :
∀ (i : Fin model.size),
v i = rhs i + if model.kind i = Spec.FiniteModel.StateKind.transient then ∑ j : Fin model.size, ↑(model.transition i j) * v j
else 0)
(hw :
∀ (i : Fin model.size),
w i = rhs i + if model.kind i = Spec.FiniteModel.StateKind.transient then ∑ j : Fin model.size, ↑(model.transition i j) * w j
else 0)
(i : Fin model.size)
:
theorem
Determinize.Proof.RewardModel.momentAt_eq_values
(model : Spec.RewardModel.Model)
(paths : FiniteModel.Paths model.control)
(pathsValid : FiniteModel.Paths.Valid model.control paths)
(values : FiniteModel.Moment → Fin model.size → ℚ)
(valid : MomentEquations model values)
(moment : FiniteModel.Moment)
(i : Fin model.size)
:
theorem
Determinize.Proof.RewardModel.integrable_outputMeasure
(model : Spec.RewardModel.Model)
:
MeasureTheory.Integrable (fun (x : ℝ) => x) model.outputMeasure ∧ MeasureTheory.Integrable (fun (x : ℝ) => x ^ 2) model.outputMeasure
theorem
Determinize.Proof.RewardModel.solution_statistics
(model : Spec.RewardModel.Model)
(solution : Finite.Reward.Solution model)
:
solution.statistics.Matches model.outputMeasure
theorem
Determinize.Proof.RewardModel.solution_result
(model : Spec.RewardModel.Model)
(program : Spec.Paper.Expr)
(matching : model.Matches program)
(solution : Finite.Reward.Solution model)
:
Spec.RewardModel.ResultMatches model program solution.statistics
theorem
Determinize.Proof.RewardModel.solution_conditional_variance
(model : Spec.RewardModel.Model)
(solution : Finite.Reward.Solution model)
(positive : 0 < solution.statistics.returnMass)
:
ProbabilityTheory.variance id ((model.outputMeasure Set.univ)⁻¹ • model.outputMeasure) = ↑(solution.statistics.secondMoment / solution.statistics.returnMass - (solution.statistics.firstMoment / solution.statistics.returnMass) ^ 2)
theorem
Determinize.Proof.RewardModel.solution_termination
(model : Spec.RewardModel.Model)
(solution : Finite.Reward.Solution model)
: