math / data /lean /OAI /Probability /MemoryPrecision /CubeSuccess.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
4.21 kB
import Mathlib
import OAI.Probability.GaussianRegression.Main
import OAI.Probability.MemoryPrecision.CubeArithmetic
import OAI.Probability.MemoryPrecision.KernelCube
import OAI.Probability.MemoryPrecision.Precision
namespace OAI
open MeasureTheory ProbabilityTheory Set Filter EuclideanGeometry
open scoped ENNReal NNReal RealInnerProductSpace Topology Pointwise
noncomputable section
universe u v
open MeasureTheory ProbabilityTheory Set Filter
open scoped ENNReal NNReal RealInnerProductSpace Topology Pointwise
namespace MemoryPrecision.Kernel
open MemoryPrecision NoiselessRegression Policy
lemma source_cube_bound {d M T : ℕ} {Ω : Type u} [MeasurableSpace Ω]
(L : Ω → Policy d M T) (ρ : Measure Ω) [IsProbabilityMeasure ρ]
(hd : 16 ≤ d) (hL : Admissible L ρ (CubePrior (by omega)))
(hM : M ≤ d^2) {ε : ℝ} (hε : 0 < ε) :
Success L ρ (CubePrior (by omega)) ε ≤ min 1 (ENNReal.ofReal
(CubeSuccessConstant^((d/2)*(BlockCount T ((d/2)/4)+1)) * ε^((d/2:ℕ)/4:ℝ))) := by
refine le_min (Policy.success_le_one L ρ (CubePrior (by omega)) ε) ?_
have hb := (Policy.success_le_chord L ρ (CubePrior (by omega)) ε).trans
(Policy.source_cube_chord_bound L ρ hd hL hε)
have he : (T+1:ℝ≥0∞) * ENNReal.ofReal
((12:ℝ)^(d/2)*ε^((d/2:ℕ)/4:ℝ) * CubeStep d (2^M)^MemoryPrecision.BlockCount T ((d/2)/4)) =
ENNReal.ofReal ((T+1:ℝ) *
((12:ℝ)^(d/2)*ε^((d/2:ℕ)/4:ℝ) * CubeStep d (2^M)^MemoryPrecision.BlockCount T ((d/2)/4))) := by
rw [ENNReal.ofReal_mul (by positivity : (0:ℝ) ≤ T+1)]
congr 1
simp [ENNReal.ofReal_add]
rw [he] at hb
apply hb.trans (ENNReal.ofReal_le_ofReal ?_)
calc
_ = ((T+1:ℝ) * ((12:ℝ)^(d/2) * CubeStep d (2^M)^MemoryPrecision.BlockCount T ((d/2)/4))) *
ε^((d/2:ℕ)/4:ℝ) := by ring
_ ≤ _ := mul_le_mul_of_nonneg_right (cube_whole_factor hd hM T) (by positivity)
lemma cubePrior_dimension {d₀ d : ℕ} (hd₀ : 16 ≤ d₀) (hd : d₀ ≤ d) : 2 ≤ d :=
Nat.le_trans (by decide : 2 ≤ 16) (Nat.le_trans hd₀ hd)
def CubeSuccessEstimate {d M T : ℕ} {Ω : Type u} [MeasurableSpace Ω]
(D : ℝ) (hd : 2 ≤ d) (ε : ℝ) (K : Ω → Policy d M T) (ρ : Measure Ω) : Prop :=
Policy.Success K ρ (CubePrior hd) ε ≤ min 1 (ENNReal.ofReal
(D^((d/2)*(MemoryPrecision.Kernel.BlockCount T ((d/2)/4)+1)) * ε^((d/2:ℕ)/4:ℝ)))
def PointwisePrecisionEstimate {d M T : ℕ} {Ω : Type u} [MeasurableSpace Ω]
(c ε : ℝ) (K : Ω → Policy d M T) (ρ : Measure Ω) : Prop :=
(∀ signal : MemoryPrecision.Kernel.Sphere d,
(@OfNat.ofNat ℝ≥0∞ 2
(@instOfNatAtLeastTwo ℝ≥0∞ 2 ENNReal.instAddCommMonoidWithOne.toNatCast inferInstance)) /
(@OfNat.ofNat ℝ≥0∞ 3
(@instOfNatAtLeastTwo ℝ≥0∞ 3 ENNReal.instAddCommMonoidWithOne.toNatCast inferInstance)) ≤
Policy.FixedSuccess K ρ ε signal) → c*(d:ℝ)*Real.log (1/ε) ≤ (T:ℝ)
abbrev SourceCubeStatement : Prop :=
∃ D : ℝ, Real.exp 1 ≤ D ∧ ∃ d₀ : ℕ, ∃ hd₀ : 16 ≤ d₀, ∃ c : ℝ, 0 < c ∧
∀ d M T : ℕ, ∀ hd : d₀ ≤ d, M ≤ d^2 → ∀ ε : ℝ, 0 < ε → ε ≤ 1/10 →
∀ (Ω : Type u) [MeasurableSpace Ω] (ρ : Measure Ω), IsProbabilityMeasure ρ →
∀ K : Ω → MemoryPrecision.Kernel.Policy d M T, Policy.ModelAdmissible.{u} K ρ →
CubeSuccessEstimate.{u} D (cubePrior_dimension hd₀ hd) ε K ρ ∧
PointwisePrecisionEstimate.{u} c ε K ρ
theorem source_cube_success : SourceCubeStatement := by
refine ⟨CubeSuccessConstant,cubeSuccessConstant_exp,32,by norm_num,
1/(8+20*StreamConstant*(1+1)),by have := streamConstant_pos; positivity,?_⟩
intro d M T hd hM ε hε hε1 Ω _ ρ hρ L hL
have : IsProbabilityMeasure ρ := hρ
let : NeZero d := ⟨by omega⟩
refine ⟨source_cube_bound L ρ (by omega) (hL _ inferInstance) hM hε, ?_⟩
intro hs
apply Policy.source_precision L ρ hd (hL _ inferInstance) (by norm_num : (0:ℝ) < 1)
(by exact_mod_cast (by simpa using hM : M ≤ 1*d^2)) hε hε1
exact Policy.source_uniform_of_pointwise L ρ (by omega) (hL _ inferInstance) ε hs
end MemoryPrecision.Kernel
end
end OAI