File size: 5,108 Bytes
dd14c26 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 | import Mathlib
import OAI.Probability.LogConcave.OraclePrograms.RoutineKind
import OAI.Probability.LogConcave.Numerics.CompileDeclaredRootShift
import OAI.Probability.LogConcave.OraclePrograms.TerminalMeanProgram
import OAI.Probability.LogConcave.Sampling.SampleCorrelation
namespace OAI
section
noncomputable section
namespace LogConcaveSampling.OracleCompiler
open MeanTree Quadrature
open scoped Classical
structure CircuitParameters where
n : ℕ
m : ℕ
N : ℕ
nc : ℕ
Nc : ℕ
T : ℝ
h : ℝ
ψ : ℝ
v : ℝ
D : ℝ
A : ℝ
Af : ℝ
T_lower : 1/2≤T
T_upper : T<1
h_pos : 0<h
v_nonneg : 0≤v
angle_bound : ∀j : Fin (m+1),|ψ*angleNodes m j|≤v
angle_small : v^2≤Real.sqrt (1-T^2)
namespace CircuitParameters
variable (C : CircuitParameters) {d : ℕ}
def meanEndpoint : ProbabilityNode C.T C.h (C.n+1) :=
probabilityEndpoint (by linarith [C.T_lower]) C.T_upper C.h_pos (C.n+1)
def meanTree (r σ : ℝ) : MeanTree (MeanInput d) d :=
literalMean r C.T C.h C.ψ σ C.n C.m C.N C.nc C.Nc C.meanEndpoint
def meanDrift (r σ : ℝ) (hr : 0<r) : DriftData (C.meanTree (d:=d) r σ) :=
(literalMean_drift hr C.T_lower C.T_upper C.h_pos C.v_nonneg C.angle_small C.angle_bound C.meanEndpoint).choose
lemma meanDrift_spec (r σ : ℝ) (hr : 0<r) :
DriftValid (meanInputMove r C.T) (C.meanTree (d:=d) r σ) (C.meanDrift r σ hr) ∧
(C.meanDrift (d:=d) r σ hr).root _=0 ∧ DriftBound 3 _ (C.meanDrift (d:=d) r σ hr) :=
(literalMean_drift hr C.T_lower C.T_upper C.h_pos C.v_nonneg C.angle_small C.angle_bound C.meanEndpoint).choose_spec
def sampleEndpoint (η : ℝ) (hη : 0<η) (hη1 : η≤1) :
ProbabilityNode (sampleCorrelation η) C.h (C.n+1) :=
probabilityEndpoint (by linarith [(sampleCorrelation_properties hη hη1).1])
(sampleCorrelation_properties hη hη1).2.1 C.h_pos (C.n+1)
def sampleTree (r η : ℝ) (hη : 0<η) (hη1 : η≤1) : MeanTree (SampleInput d) d :=
literalSample r (sampleCorrelation η) C.h (C.n+1) C.N (C.sampleEndpoint η hη hη1)
def sampleDrift (r η : ℝ) (hr : 0<r) (hη : 0<η) (hη1 : η≤1) :
DriftData (C.sampleTree (d:=d) r η hη hη1) :=
(literalSample_drift hr (Nat.succ_pos C.n)
(by linarith [(sampleCorrelation_properties hη hη1).1])
(sampleCorrelation_properties hη hη1).2.1 C.h_pos (C.sampleEndpoint η hη hη1)).choose
def meanCenterBudget (r σ : ℝ) : ℝ := C.A*(r/σ+r^2)
def sampleCenterBudget (r : ℝ) : ℝ := C.A*r^2
def sampleFinalBudget (r : ℝ) : ℝ := C.Af*r
def meanFrom (Mf Mi : ℝ → ℝ → SeedProgram d) (S : ℝ → ℝ → ReservedProgram d)
(r σ : ℝ) : SeedProgram d :=
if hr : 0<r then
let E := C.meanTree r σ
let p := C.meanDrift r σ hr
let noise := Real.sqrt 3*σ/2
let B := compileDeclaredRoot C.D (C.meanCenterBudget r σ) C.Af Mf Mi E noise
let v := compileDeclaredRootShift C.D (C.meanCenterBudget r σ) C.Af Mf Mi E noise p
let R := Real.sqrt (1-C.T^2)
let anchor := S r (R/C.T)
meanParent anchor.full B (anchor.extraShift (-2*R/C.T)) v C.T
else terminalMeanProgram d r σ
def sampleFrom (Mf Mi : ℝ → ℝ → SeedProgram d) (r η : ℝ) : ReservedProgram d :=
if hv : 0<r ∧ 0<η ∧ η≤1 then
let E := C.sampleTree r η hv.2.1 hv.2.2
let p := C.sampleDrift r η hv.1 hv.2.1 hv.2.2
let noise := η/Real.sqrt 2
let B := compileDeclaredRoot C.D (C.sampleCenterBudget r) (C.sampleFinalBudget r) Mf Mi E noise
let v := compileDeclaredRootShift C.D (C.sampleCenterBudget r) (C.sampleFinalBudget r) Mf Mi E noise p
sampleParent B v (sampleCorrelation η) η
else terminalReservedProgram d r η
end CircuitParameters
def RoutineValue (d : ℕ) : RoutineKind → Type
| .mean => ℝ → ℝ → SeedProgram d
| .sample => ℝ → ℝ → ReservedProgram d
def constructRoutine (d : ℕ) (C : CircuitParameters) {t : ℝ} (ht : 0<t) (hs : t≤1/44)
(L : RoutineLabel) : RoutineValue d L.kind :=
if _ : L.terminal then
match L.kind with
| .mean => terminalMeanProgram d
| .sample => terminalReservedProgram d
else
match hk : L.kind with
| .mean => C.meanFrom
(constructRoutine d C ht hs ⟨.mean,L.b+t,L.p⟩)
(constructRoutine d C ht hs ⟨.mean,L.b-10*t,L.p-1/4⟩)
(constructRoutine d C ht hs ⟨.sample,L.b-10*t,L.p-1/4⟩)
| .sample => C.sampleFrom
(constructRoutine d C ht hs ⟨.mean,L.b,L.p⟩)
(constructRoutine d C ht hs ⟨.mean,L.b-10*t,L.p-1/4⟩)
termination_by L.rank t
decreasing_by
all_goals apply RoutineStep.rank_lt ht hs (by assumption)
· simpa only [←hk] using (RoutineStep.meanFinal (t:=t) L.b L.p)
· simpa only [←hk] using (RoutineStep.internal (t:=t) RoutineKind.mean RoutineKind.mean L.b L.p)
· simpa only [←hk] using (RoutineStep.internal (t:=t) RoutineKind.mean RoutineKind.sample L.b L.p)
· simpa only [←hk] using (RoutineStep.sampleFinal (t:=t) L.b L.p)
· simpa only [←hk] using (RoutineStep.internal (t:=t) RoutineKind.sample RoutineKind.mean L.b L.p)
end LogConcaveSampling.OracleCompiler
end
end
end OAI
|