Expectation and variance of returned values #
The output law bigStepMeasure program is unnormalized: its total mass, returnProbability, is
the probability that the program returns a real. extendedExpectation is the expectation of a
law in the extended reals, which is well-defined (HasExpectation) as soon as one of its positive
and negative parts is finite; it covers programs such as 1 / uniform(0, 1) whose expectation is
+∞. returnedExpectation and returnedVariance are the paper's 𝔼_ret[e] and Var_ret[e]:
the expectation and the variance of the output conditioned on returning.
The positive part ∫ v⁺ dμ of the expectation of a real law.
Equations
- Determinize.Spec.posPartIntegral μ = ∫⁻ (value : ℝ), ENNReal.ofReal value ∂μ
Instances For
The negative part ∫ v⁻ dμ of the expectation of a real law.
Equations
- Determinize.Spec.negPartIntegral μ = ∫⁻ (value : ℝ), ENNReal.ofReal (-value) ∂μ
Instances For
The expectation of μ is well-defined in [-∞, ∞] when at least one part is finite.
Equations
Instances For
The expectation ∫ v⁺ dμ - ∫ v⁻ dμ in the extended reals; meaningful under
HasExpectation μ.
Equations
Instances For
The probability q_e that the program returns a real: the total mass of its output law.
Equations
- Determinize.Spec.returnProbability program = (Determinize.Spec.Paper.bigStepMeasure program) Set.univ
Instances For
Output law conditioned on returning a real; used only when the return probability is positive.
Equations
- Determinize.Spec.returnedLaw program = (Determinize.Spec.returnProbability program)⁻¹ • Determinize.Spec.Paper.bigStepMeasure program
Instances For
The expectation 𝔼_ret[e] of the output conditioned on returning, in the extended reals:
the expectation of the output law divided by the return probability. It is meaningful when the
return probability is positive and HasExpectation (bigStepMeasure program).
Equations
Instances For
The variance Var_ret[e] of the output conditioned on returning: Mathlib's variance
(∫ (v - ∫ v)²) of the returned law.