More type stabilization; use of SnoopCompile.write to precompile

.gitignore

@@ -7,3 +7,5 @@ voters.dta
 test/profiling.jl
 src/WRE2.jl
 docs/build/
+profile.pb.gz
+src/precompiler.jl

Project.toml

@@ -1,7 +1,7 @@
 name = "WildBootTests"
 uuid = "65c2e505-86ba-4c19-93f1-95506c1443d5"
 authors = ["droodman <[email protected]>"]
-version = "0.6.2"
+version = "0.6.3"
 
 [deps]
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"

precompiler.jl

@@ -0,0 +1,80 @@
+# run this in a fresh Julia session after having commented out the precompile lines at the bottom of WildBootTests.jl, both to avoid complications and force re-precompile
+push!(LOAD_PATH, ".")
+using WildBootTests
+
+using SnoopCompileCore
+tinf = @snoopi_deep begin
+  for T in (Float32, Float64)
+    resp, predexog, clustid = rand(T, 400), rand(T, 400, 4), Int32.(rand(1900:1920, 400))
+    test = wildboottest(T, [0 0 0 1], [.04]; resp, predexog, clustid, auxwttype=WildBootTests.webb)
+    teststat(test); p(test); CI(test); plotpoints(test);
+
+    test = wildboottest(T, [0 0 0 1], [.04]; resp, predexog, clustid, reps=9999999, auxwttype=WildBootTests.webb, getCI=false)
+    teststat(test); p(test);
+
+    test = wildboottest(T, [0 0 0 1; 0 0 1 0], [0.05; -0.02]; resp, predexog, clustid, reps=9999, auxwttype=WildBootTests.webb)
+    teststat(test); p(test);
+
+    resp, predexog, clustid = rand(T, 2000), rand(T, 2000, 4), Int32.(rand(1:12, 2000))
+    test = wildboottest(T, [0 1 0 0], [0]; R1=[0 0 1 0], r1=[.2], resp, predexog, clustid)
+    teststat(test); p(test); CI(test);
+
+    resp, predexog, clustid, predendog, inst = rand(T, 2000), rand(T, 2000, 3), Int32.(rand(1:12, 2000)), rand(2000), rand(2000)
+    test = wildboottest(T, [0 0 0 1], [0]; resp, predexog, predendog, inst, clustid, small=false, reps=9999, ptype=WildBootTests.equaltail)
+    teststat(test); p(test); CI(test);
+
+    test = wildboottest(T, [0 0 0 1], [.0]; resp, predexog, predendog, inst, clustid, small=false, reps=9999, auxwttype=WildBootTests.webb, bootstrapc=true, ptype=WildBootTests.equaltail)
+    teststat(test); p(test); CI(test);
+
+    test = wildboottest(T, [0 0 0 1], [0]; resp, predexog, predendog, inst, clustid, small=false, ARubin=true, reps=9999)
+    teststat(test); p(test); CI(test);
+
+    test = wildboottest(T, [0 0 0 1], [0]; resp, predexog, predendog, inst, clustid, small=false, ARubin=true, reps=9999, imposenull=false)
+    teststat(test); p(test); CI(test);
+
+    test = wildboottest(T, [0 0 0 1], [0]; resp, predexog, predendog, inst, clustid, small=false, reps=0)
+    teststat(test); p(test); CI(test);
+
+    test = wildboottest(T, [0 0 0 1], [0]; resp, predexog, predendog, inst, clustid, small=false, reps=0, imposenull=false)
+    teststat(test); p(test); CI(test);
+
+    resp, predexog, clustid, predendog, inst = rand(T, 2000), rand(T, 2000, 1), Int32.(rand(1:12, 2000)), rand(2000), rand(2000,2)
+    test = wildboottest(T, [0 1], [0]; resp, predexog, predendog, inst, LIML=true, clustid, small=false, reps=999)
+    teststat(test); p(test); CI(test);
+
+    resp, predexog, clustid, predendog, inst = rand(T, 2000,), rand(T, 2000, 5), Int32.(rand(1:12, 2000)), rand(2000), rand(2000,2)
+    test = wildboottest(T, [0 0 0 0 0 1], [0]; resp, predexog, predendog, inst, Fuller=1, clustid, small=false, reps=9999, auxwttype=WildBootTests.webb)
+    teststat(test); p(test); CI(test);
+
+    resp, predexog, clustid, obswt, feid = rand(T, 2000,), rand(T, 2000, 3), Int32.(rand(1:12, 2000, 2)), rand(2000), Int64.(rand(1:12,2000))
+    test = wildboottest(T, [0 0 1], [0]; resp, predexog, clustid, nbootclustvar=1, nerrclustvar=2, obswt, feid)
+    teststat(test); p(test); CI(test);
+
+    resp, predexog, clustid, feid = rand(T, 20000,), rand(T, 20000, 12), Int32.(rand(1:12, 20000, 2)), Int32.(rand(1:12,20000))
+    test = wildboottest(T, [1 zeros(1,size(predexog,2)-1)], [0]; resp, predexog, clustid, nbootclustvar=1, nerrclustvar=2, feid, reps=9999)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [1 zeros(1,size(predexog,2)-1)], [0]; resp, predexog, clustid, nbootclustvar=1, nerrclustvar=2, feid, reps=9999)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [1 zeros(1,size(predexog,2)-1)], [0]; resp, predexog, clustid, nbootclustvar=2, nerrclustvar=2, feid, reps=9999)
+    teststat(test); p(test); CI(test);
+
+
+    resp, predexog, clustid = rand(T, 20000,), rand(T, 20000, 12), Int32.(rand(1:12, 20000, 2))
+    test = wildboottest(T, [0 1 zeros(1,size(predexog,2)-2)], [0]; resp, predexog, clustid=clustid[:,1], gridpoints=[10], reps=9999)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [0 1 zeros(1,size(predexog,2)-2)], [0]; resp, predexog, clustid=clustid[:,1], reps=9999, imposenull=false)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [0 1 zeros(1,size(predexog,2)-2)], [0]; resp, predexog, clustid=clustid[:,[2,1]], nbootclustvar=2, nerrclustvar=1, reps=9999)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [0 1 zeros(1,size(predexog,2)-2)], [0]; resp, predexog, clustid=clustid[:,[2,1]], nbootclustvar=2, nerrclustvar=1, reps=9999, imposenull=false)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [0 1 zeros(1,size(predexog,2)-2)], [0]; resp, predexog, clustid= [collect(1:size(resp,1)) clustid[:,2]], nbootclustvar=1, nerrclustvar=1, reps=9999)
+    teststat(test); p(test); CI(test);
+    test = wildboottest(T, [0 1 zeros(1,size(predexog,2)-2)], [0]; resp, predexog, clustid= [collect(1:size(resp,1)) clustid[:,2]], nbootclustvar=1, nerrclustvar=1, reps=9999, imposenull=false)
+    teststat(test); p(test); CI(test);
+  end
+end
+
+using SnoopCompile
+ttot, pcs = SnoopCompile.parcel(tinf);
+SnoopCompile.write("/src", [pcs[end]])

src/StrBoottest.jl

@@ -1,4 +1,4 @@
-# Definition of StrBoottest "class" for holding intermediate results, with associated utilities and get functions
+# Definition of StrBootTest "class" for holding intermediate results, with associated utilities and get functions
 
 "Auxilliary weight types: `rademacher`, `mammen`, `webb`, `normal`, `gamma`"
 @enum AuxWtType rademacher mammen webb normal gamma
@@ -23,9 +23,35 @@ struct StrFE{T<:Real}
   wt::Vector{T}
 end
 
+mutable struct StrEstimator{T<:AbstractFloat}
+  isDGP::Bool; LIML::Bool; Fuller::T; κ::T
+  R₁perp::Matrix{T}; Rpar::Matrix{T}
 
-@inline vecconvert(T::DataType, X) = isa(X, AbstractArray) ? reshape(eltype(X)==T ? X : T.(X), size(X,1)           ) : X
-@inline matconvert(T::DataType, X) = isa(X, AbstractArray) ? reshape(eltype(X)==T ? X : T.(X), size(X,1), size(X,2)) : X
+  kZ::Int64
+  y₁::Vector{T}; ü₁::Vector{T}; u⃛₁::Vector{T}; β̂::Vector{T}; β̂₀::Vector{T}; invXXXy₁par::Vector{T}
+  Yendog::Vector{Bool}
+  invZperpZperp::Symmetric{T,Matrix{T}}; XZ::Matrix{T}; PXZ::Matrix{T}; YPXY::Symmetric{T,Matrix{T}}; R₁invR₁R₁::Matrix{T}
+	RperpX::#=Uber=#Matrix{T}; RperpXperp::#=Uber=#Matrix{T}; RRpar::Matrix{T}; RparY::#=Uber=#Matrix{T}; RR₁invR₁R₁::Matrix{T}
+	∂β̂∂r::Matrix{T}; YY::Symmetric{T,Matrix{T}}; AR::Matrix{T}; XAR::Matrix{T}; R₁invR₁R₁Y::Matrix{T}; invXXXZ::Matrix{T}; Ü₂::Matrix{T}; XinvXX::Matrix{T}; Rt₁::Vector{T}
+	invXX::Symmetric{T,Matrix{T}}; Y₂::Matrix{T}; X₂::Matrix{T}; invH::Symmetric{T,Matrix{T}}
+	y₁par::Vector{T}; Xy₁par::Vector{T}
+	A::Symmetric{T,Matrix{T}}; Z::Matrix{T}; Zperp::Matrix{T}; X₁::Matrix{T}
+	FillingT₀::Matrix{Matrix{T}}
+	WXAR::Matrix{T}; S⋂PXYZperp::Vector{Matrix{T}}; S⋂YX::Vector{Matrix{T}}; CT_XAR::Vector{Matrix{T}}; CT_FE⋂PY::Vector{Matrix{T}}
+
+  # IV/GMM only
+  ZZ::Symmetric{T,Matrix{T}}; XY₂::Matrix{T}; XX::Symmetric{T,Matrix{T}}; H_2SLS::Symmetric{T,Matrix{T}}; V::Matrix{T}; ZY₂::Matrix{T}; X₂Y₂::Matrix{T}; X₁Y₂::Matrix{T}; ZR₁ZR₁::Symmetric{T,Matrix{T}}; X₂ZR₁::Matrix{T}; ZR₁Y₂::Matrix{T}; X₁ZR₁::Matrix{T}
+  ZZR₁::Matrix{T}; X₂y₁::Vector{T}; X₁y₁::Vector{T}; Zy₁::Vector{T}; ZXinvXXXZ::Matrix{T}; H_2SLSmZZ::Symmetric{T,Matrix{T}}
+  ZXinvXXXy₁par::Vector{T}; t₁Y::Vector{T}
+  Y₂y₁::Vector{T}; twoR₁Zy₁::Vector{T}
+  y₁y₁::T; y₁pary₁par::T
+  X₂y₁par::Vector{T}; X₁y₁par::Vector{T}; Zy₁par::Vector{T}
+  Y₂y₁par::Vector{T}
+  Rperp::Matrix{T}; ZR₁::Matrix{T}
+  kX::Int64
+
+  StrEstimator{T}(isDGP, LIML, Fuller, κ) where T<:AbstractFloat = new(isDGP, LIML, Fuller, κ, Matrix{T}(undef,0,0))
+end
 
 mutable struct StrBootTest{T<:AbstractFloat}
   R::Matrix{T}; r::Vector{T}; R₁::Matrix{T}; r₁::Vector{T}
@@ -34,11 +60,11 @@ mutable struct StrBootTest{T<:AbstractFloat}
   LIML::Bool; Fuller::T; κ::T; ARubin::Bool
   B::Int64; auxtwtype::AuxWtType; rng::AbstractRNG; maxmatsize::Float16
   ptype::PType; null::Bool; bootstrapt::Bool
-	ID::Matrix{Int64}; nbootclustvar::Int8; nerrclustvar::Int64; issorted::Bool; small::Bool
+	ID::Matrix{Int64}; nbootclustvar::Int8; nerrclustvar::Int8; issorted::Bool; small::Bool
   FEID::Vector{Int64}; FEdfadj::Bool
   level::T; rtol::T
   madjtype::MAdjType; NH₀::Int16
-  ML::Bool; β̂::Vector{T}; A::Matrix{T}; sc::Matrix{T}
+  ML::Bool; β̂::Vector{T}; A::Symmetric{T,Matrix{T}}; sc::Matrix{T}
   willplot::Bool; gridmin::Vector{T}; gridmax::Vector{T}; gridpoints::Vector{Float32}
 
   q::Int16; twotailed::Bool; scorebs::Bool; robust::Bool
@@ -52,14 +78,14 @@ mutable struct StrBootTest{T<:AbstractFloat}
   peak::NamedTuple{(:X, :p), Tuple{Vector{T}, T}}
 
   sqrt::Bool; Nobs::Int64; _Nobs::T; kZ::Int64; kY₂::Int64; kX₁::Int64; sumwt::T; NClustVar::Int8; haswt::Bool; REst::Bool; multiplier::T; smallsample::T
-		WREnonARubin::Bool; dof::Int64; dof_r::Int64; p::T; BootClust::Int8
+		WREnonARubin::Bool; dof::Int64; dof_r::T; p::T; BootClust::Int8
 		purerobust::Bool; N✻::Int64; Nw::Int64; enumerate::Bool; interpolable::Bool; interpolate_u::Bool; kX₂::Int64; kX::Int64
   _FEID::Vector{Int64}; AR::Matrix{T}; v::Matrix{T}; u✻::Matrix{T}; CT_WE::Matrix{T}
   infoBootData::Vector{UnitRange{Int64}}; infoBootAll::Vector{UnitRange{Int64}}; infoErrAll::Vector{UnitRange{Int64}}
   JN⋂N✻::Matrix{T}; statDenom::Matrix{T}; uXAR::Matrix{T}; SuwtXA::Matrix{T}; numer₀::Matrix{T}; β̂dev::Matrix{T}; δdenom_b::Matrix{T}
 	_J⋂::Matrix{T}; YY✻_b::Matrix{T}; YPXY✻_b::Matrix{T}; numerw::Matrix{T}; Zyg::Vector{Matrix{T}}; numer_b::Vector{T}
 
-	distCDR::Matrix{T}; plotX::Tuple{Vararg{Vector{T}, N} where N}; plotY::Vector{T}; ClustShare::Vector{T}; WeightGrp::Vector{UnitRange{Int64}}
+	distCDR::Matrix{T}; plotX::Vector{Vector{T}}; plotY::Vector{T}; ClustShare::Vector{T}; WeightGrp::Vector{UnitRange{Int64}}
   numersum::Vector{T}; ü₀::Vector{T}; invFEwt::Vector{T}
 	β̂s::Matrix{T}; As::Matrix{T}
 	infoAllData::Vector{UnitRange{Int64}}; info⋂Data::Vector{UnitRange{Int64}}; IDAll::Matrix{T}
@@ -78,9 +104,9 @@ mutable struct StrBootTest{T<:AbstractFloat}
   StrBootTest{T}(R, r, R₁, r₁, y₁, X₁, Y₂, X₂, wt, fweights, LIML, 
 	               Fuller, κ, ARubin, B, auxtwtype, rng, maxmatsize, ptype, null, scorebs, bootstrapt, ID, nbootclustvar, nerrclustvar, issorted, robust, small, FEID, FEdfadj, level, rtol, madjtype, NH₀, ML,
 								 β̂, A, sc, willplot, gridmin, gridmax, gridpoints) where T<:Real =
-	  new(matconvert(T,R), vecconvert(T,r), matconvert(T,R₁), vecconvert(T,r₁), vecconvert(T,y₁), matconvert(T,X₁), matconvert(T,Y₂), matconvert(T,X₂), vecconvert(T,wt), fweights, LIML || !iszero(Fuller), 
-		    Fuller, κ, ARubin, B, auxtwtype, rng, maxmatsize, ptype, null, bootstrapt, matconvert(Int64,ID), nbootclustvar, nerrclustvar, issorted, small, vecconvert(Int64,FEID), FEdfadj, level, rtol, madjtype, NH₀, ML, 
-				vecconvert(T,β̂), matconvert(T,A), matconvert(T,sc), willplot, gridmin, gridmax, gridpoints,
+	  new(R, r, R₁, r₁, y₁, X₁, Y₂, X₂, wt, fweights, LIML || !iszero(Fuller), 
+		    Fuller, κ, ARubin, B, auxtwtype, rng, maxmatsize, ptype, null, bootstrapt, ID, nbootclustvar, nerrclustvar, issorted, small, FEID, FEdfadj, level, rtol, madjtype, NH₀, ML, 
+				β̂, A, sc, willplot, gridmin, gridmax, gridpoints,
 		  nrows(R),
 		  ptype==symmetric || ptype==equaltail,
 		  scorebs || iszero(B) || ML,
@@ -92,7 +118,6 @@ mutable struct StrBootTest{T<:AbstractFloat}
 		  Vector{T}(undef,0), Vector{T}(undef,0), Matrix{T}(undef,0,0), Vector{T}(undef,0),
 		  Matrix{T}(undef,0,0),
 		  (X = Vector{T}(undef,0), p = T(NaN)))
-
 end
 
 
@@ -334,7 +359,7 @@ function getdf_r(o::StrBootTest)
 end
 function _getplot(o::StrBootTest)
   o.notplotted && plot!(o)
-  (X=o.plotX, p=o.plotY)
+  (X=Tuple(o.plotX), p=o.plotY)
 end
 function getpeak(o::StrBootTest)  # x and y values of confidence curve peak (at least in OLS && ARubin)
   o.notplotted && plot!(o)

src/WRE.jl

@@ -92,12 +92,12 @@ function Filling(o::StrBootTest{T}, ind1::Integer, β̂s::AbstractMatrix) where
 			PXY✻ = reshape(o.Repl.PXZ[:,ind1], :, 1)  # store as matrix to reduce compiler confusion
 			o.Repl.Yendog[ind1+1] && (PXY✻ = PXY✻ .+ o.S✻UPX[ind1+1] * o.v)
 
-			dest = @panelsum(PXY✻ .* (o.Repl.y₁ .- o.S✻UMZperp[1] * o.v), o.wt, o.info⋂Data)
+			dest = @panelsum(PXY✻ .* (o.Repl.y₁ .- o.S✻UMZperp[1] * o.v), o.info⋂Data)
 
 			for ind2 ∈ 1:o.Repl.kZ
 				_β̂ = view(β̂s,ind2,:)'
 				dest .-= @panelsum(PXY✻ .* (o.Repl.Yendog[ind2+1] ?  view(o.Repl.Z,:,ind2) * _β̂ .- o.S✻UMZperp[ind2+1] * (o.v .* _β̂) :
-															                              (view(o.Repl.Z,:,ind2) * _β̂)                                      ), o.wt, o.info⋂Data)
+															                              (view(o.Repl.Z,:,ind2) * _β̂)                                      ), o.info⋂Data)
 			end
 		else  # create pieces of each N x B matrix one at a time rather than whole thing at once
 			dest = Matrix{T}(undef, o.clust[1].N, ncols(o.v))  # XXX preallocate this & turn Filling into Filling! ?
@@ -110,11 +110,11 @@ function Filling(o::StrBootTest{T}, ind1::Integer, β̂s::AbstractMatrix) where
 						o.Repl.Yendog[ind1+1] && (PXY✻ = PXY✻ .+ view(o.S✻UPX[ind1+1],i,:)'o.v)
 
 						if iszero(ind2)
-							dest[i,:]   = wtsum(o.wt, PXY✻ .* (o.Repl.y₁[i] .- view(o.S✻UMZperp[1],i,:))'o.v)
+							dest[i,:]   = colsum(PXY✻ .* (o.Repl.y₁[i] .- view(o.S✻UMZperp[1],i,:))'o.v)
 						elseif o.Repl.Yendog[ind2+1]
-							dest[i,:] .-= wtsum(o.wt, PXY✻ .* (o.Repl.Z[i,ind2] * _β̂ .- view(o.S✻UMZperp[ind2+1],i,:)'β̂v))
+							dest[i,:] .-= colsum(PXY✻ .* (o.Repl.Z[i,ind2] * _β̂ .- view(o.S✻UMZperp[ind2+1],i,:)'β̂v))
 						else
-							dest[i,:] .-= wtsum(o.wt, PXY✻ .* (o.Repl.Z[i,ind2] * _β̂))
+							dest[i,:] .-= colsum(PXY✻ .* (o.Repl.Z[i,ind2] * _β̂))
 						end
 					end
 				else
@@ -124,9 +124,9 @@ function Filling(o::StrBootTest{T}, ind1::Integer, β̂s::AbstractMatrix) where
 																					 reshape(o.Repl.PXZ[S,ind1], :, 1)
 
 						if iszero(ind2)
-							dest[i,:]   = wtsum(o.wt, PXY✻ .* (o.Repl.y₁[S] .- view(o.S✻UMZperp[1],S,:) * o.v))
+							dest[i,:]   = colsum(PXY✻ .* (o.Repl.y₁[S] .- view(o.S✻UMZperp[1],S,:) * o.v))
 						else
-							dest[i,:] .-= wtsum(o.wt, PXY✻ .* (o.Repl.Yendog[ind2+1] ? o.Repl.Z[S,ind2] * _β̂ .- view(o.S✻UMZperp[ind2+1],S,:) * β̂v :
+							dest[i,:] .-= colsum(PXY✻ .* (o.Repl.Yendog[ind2+1] ? o.Repl.Z[S,ind2] * _β̂ .- view(o.S✻UMZperp[ind2+1],S,:) * β̂v :
 																						                             o.Repl.Z[S,ind2] * _β̂                                       ))
 						end
 					end
@@ -180,7 +180,7 @@ function Filling(o::StrBootTest{T}, ind1::Integer, β̂s::AbstractMatrix) where
 			if o.Repl.Yendog[ind1+1] && o.Repl.Yendog[ind2+1]
 				for i ∈ 1:o.clust[1].N
 					S = o.info⋂Data[i]
-					colquadformminus!(dest, i, cross(view(o.S✻UPX[ind1+1],S,:), o.haswt ? o.wt[S] : zeros(T,0), view(o.S✻UMZperp[ind2+1],S,:)), o.v, β̂v)
+					colquadformminus!(dest, i, view(o.S✻UPX[ind1+1],S,:)'view(o.S✻UMZperp[ind2+1],S,:), o.v, β̂v)
 				end
 			end
 		end
@@ -190,12 +190,13 @@ end
 
 
 function PrepWRE!(o::StrBootTest{T}) where T
-  EstimateIVGMM!(o.DGP, o.null ? [o.r₁ ; o.r] : o.r₁)
-  MakeResidualsIVGMM!(o.DGP)
+  EstimateIV!(o.DGP, o, o.null ? [o.r₁ ; o.r] : o.r₁)
+  MakeResidualsIV!(o.DGP, o)
   Ü₂par = view(o.DGP.Ü₂ * o.Repl.RparY,:,:)
 
   for i ∈ 0:o.Repl.kZ  # precompute various clusterwise sums
-		uwt = vHadw(i>0 ? view(Ü₂par,:,i) : view(o.DGP.u⃛₁,:), o.wt)::Union{Vector{T}, SubArray{T, 1}}
+		u = i>0 ? view(Ü₂par,:,i) : view(o.DGP.u⃛₁,:)
+		uwt = vHadw(u, o.wt)::Union{Vector{T}, SubArray{T, 1}}
 
 		# S_✻(u .* X), S_✻(u .* Zperp) for residuals u for each endog var; store transposed
 		o.S✻UX[i+1]      = @panelsum2(o.Repl.X₁, o.Repl.X₂, uwt, o.infoBootData)'
@@ -217,7 +218,7 @@ function PrepWRE!(o::StrBootTest{T}) where T
 		if o.robust && o.bootstrapt
 			if !o.granular  # Within each bootstrap cluster, groupwise sum by all-error-cluster-intersections of u.*X and u.Zperp (and times invXX or invZperpZperp)
 				for g ∈ 1:o.N✻
-					o.SCT⋂uXinvXX[i+1,g] = @panelsum(o.Repl.XinvXX, uwt, o.infoCT⋂✻[g])
+					o.SCT⋂uXinvXX[i+1,g] = @panelsum(o.Repl.XinvXX, u, o.infoCT⋂✻[g])
 				end
 			end
 

src/WildBootTests.jl

@@ -5,8 +5,8 @@ export BoottestResult, wildboottest, AuxWtType, PType, MAdjType, DistStatType,
 using LinearAlgebra, Random, Distributions, SortingAlgorithms, LoopVectorization
 
 include("utilities.jl")
-include("estimators.jl")
 include("StrBoottest.jl")
+include("estimators.jl")
 include("init.jl")
 include("WRE.jl")
 include("nonWRE.jl")
@@ -50,7 +50,7 @@ function NoNullUpdate!(o::StrBootTest{T} where T)
   if o.WREnonARubin
 		o.numer[:,1] = o.R * o.DGP.Rpar * o.β̂s[1] - o.r
   elseif o.ARubin
-		EstimateARubin!(o.DGP, o.r)
+		EstimateARubin!(o.DGP, o, o.r)
 		o.numer[:,1] = o.v_sd * @view o.DGP.β̂[o.kX₁+1:end,:]  # coefficients on excluded instruments in ARubin OLS
   else
 		o.numer[:,1] = o.v_sd * (o.R * (o.ML ? o.β̂ : o.M.β̂) - o.r)  # Analytical Wald numerator; if imposing null then numer[:,1] already equals this. If not, then it's 0 before this
@@ -80,5 +80,9 @@ function UpdateBootstrapcDenom!(o::StrBootTest{T} where T, w::Integer)
 	nothing
 end
 
-include("precompile.jl")
+if Base.VERSION >= v"1.4.2"  # source: https://timholy.github.io/SnoopCompile.jl/stable/snoopi_deep_parcel/#SnoopCompile.write
+	include("../src/precompile_WildBootTests.jl")
+	_precompile_()
+end
+
 end