Equations
- Determinize.Proof.RewardModel.additionStack offsets = List.map (fun (c : ℚ) => Determinize.Finite.Frame.right Determinize.Finite.Binary.add (Determinize.Finite.Value.number c)) offsets
Instances For
theorem
Determinize.Proof.RewardModel.numeric_addition_output
(offsets : List ℚ)
(b : ℚ)
(n : ℕ)
:
FiniteModel.machineOutput (n + offsets.length) (Finite.State.deliver (Finite.Value.number b) (additionStack offsets)) = MeasureTheory.Measure.dirac ↑(b + offsets.sum)
theorem
Determinize.Proof.RewardModel.step_terminal
(state : Finite.State)
(result : Finite.Step)
(action : Finite.step state = Except.ok result)
:
(∃ (tag : Finite.Evidence) (outcomes : List (ℚ × Finite.State)), result = Finite.Step.next tag outcomes) ∨ (∃ (b : ℚ), state = Finite.State.deliver (Finite.Value.number b) [] ∧ result = Finite.Step.returned b) ∨ state = Finite.State.rejected ∧ result = Finite.Step.rejected
theorem
Determinize.Proof.RewardModel.addition_output_lower
(offsets : List ℚ)
(n : ℕ)
(state : Finite.State)
:
Spec.RewardModel.shift offsets.sum (FiniteModel.machineOutput n state) ≤ FiniteModel.machineOutput (n + offsets.length) (FiniteModel.pushStack state (additionStack offsets))
theorem
Determinize.Proof.RewardModel.addition_failure
(state : Finite.State)
(offsets : List ℚ)
(failure : Finite.Failure)
(action : Finite.step state = Except.error failure)
:
∃ (error : Finite.Failure), Finite.step (FiniteModel.pushStack state (additionStack offsets)) = Except.error error
theorem
Determinize.Proof.RewardModel.numeric_addition_upper
(offsets : List ℚ)
(b : ℚ)
(n : ℕ)
:
FiniteModel.machineOutput n (Finite.State.deliver (Finite.Value.number b) (additionStack offsets)) ≤ MeasureTheory.Measure.dirac ↑(b + offsets.sum)
theorem
Determinize.Proof.RewardModel.addition_output_upper
(offsets : List ℚ)
(n : ℕ)
(state : Finite.State)
:
FiniteModel.machineOutput n (FiniteModel.pushStack state (additionStack offsets)) ≤ Spec.RewardModel.shift offsets.sum (FiniteModel.machineOutput n state)
theorem
Determinize.Proof.RewardModel.numeric_addition_exact
(offsets : List ℚ)
(b : ℚ)
(n : ℕ)
:
FiniteModel.machineOutput n (Finite.State.deliver (Finite.Value.number b) (additionStack offsets)) = if offsets.length ≤ n then MeasureTheory.Measure.dirac ↑(b + offsets.sum) else 0
theorem
Determinize.Proof.RewardModel.addition_guard_upper
(offsets : List ℚ)
(nonempty : offsets ≠ [])
(n : ℕ)
(state : Finite.State)
:
FiniteModel.machineOutput n (FiniteModel.pushStack state (additionStack offsets)) ≤ Spec.RewardModel.shift offsets.sum (FiniteModel.machineOutput n (FiniteModel.pushStack state (additionStack [0])))