Merge pull request #50 from JuliaML/auto-juliaformatter-pr

[AUTO] JuliaFormatter.jl run

src/DensityRatioEstimation.jl

@@ -36,25 +36,25 @@ include("kmm/julia.jl")
 using Requires
 function __init__()
   # KMM
-  @require JuMP="4076af6c-e467-56ae-b986-b466b2749572" begin
-    @require Ipopt="b6b21f68-93f8-5de0-b562-5493be1d77c9" include("kmm/jump.jl")
+  @require JuMP = "4076af6c-e467-56ae-b986-b466b2749572" begin
+    @require Ipopt = "b6b21f68-93f8-5de0-b562-5493be1d77c9" include("kmm/jump.jl")
   end
 
   # KLIEP
-  @require Optim="429524aa-4258-5aef-a3af-852621145aeb" include("kliep/optim.jl")
-  @require Convex="f65535da-76fb-5f13-bab9-19810c17039a" begin
-    @require ECOS="e2685f51-7e38-5353-a97d-a921fd2c8199" include("kliep/convex.jl")
+  @require Optim = "429524aa-4258-5aef-a3af-852621145aeb" include("kliep/optim.jl")
+  @require Convex = "f65535da-76fb-5f13-bab9-19810c17039a" begin
+    @require ECOS = "e2685f51-7e38-5353-a97d-a921fd2c8199" include("kliep/convex.jl")
   end
 
   # LSIF
-  @require Optim="429524aa-4258-5aef-a3af-852621145aeb" include("lsif/optim.jl")
-  @require JuMP="4076af6c-e467-56ae-b986-b466b2749572" begin
-    @require Ipopt="b6b21f68-93f8-5de0-b562-5493be1d77c9" include("lsif/jump.jl")
+  @require Optim = "429524aa-4258-5aef-a3af-852621145aeb" include("lsif/optim.jl")
+  @require JuMP = "4076af6c-e467-56ae-b986-b466b2749572" begin
+    @require Ipopt = "b6b21f68-93f8-5de0-b562-5493be1d77c9" include("lsif/jump.jl")
   end
 
   # AD and GPU libs
-  @require Zygote="e88e6eb3-aa80-5325-afca-941959d7151f" include("utils/zygote.jl")
-  @require CuArrays="3a865a2d-5b23-5a0f-bc46-62713ec82fae" include("utils/cuarrays.jl")
+  @require Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f" include("utils/zygote.jl")
+  @require CuArrays = "3a865a2d-5b23-5a0f-bc46-62713ec82fae" include("utils/cuarrays.jl")
 end
 
 export
@@ -67,7 +67,10 @@ export
 
   # estimators
   DensityRatioEstimator,
-  KMM, uKMM, KLIEP, LSIF,
+  KMM,
+  uKMM,
+  KLIEP,
+  LSIF,
   available_optlib,
   default_optlib,
   densratiofunc,

src/api/estimators.jl

@@ -26,8 +26,7 @@ the list below:
 
 See also [`densratiofunc`](@ref).
 """
-densratio(x_nu, x_de, dre::DensityRatioEstimator;
-          optlib=default_optlib(typeof(dre))) =
+densratio(x_nu, x_de, dre::DensityRatioEstimator; optlib=default_optlib(typeof(dre))) =
   _densratio(x_nu, x_de, dre, optlib)
 
 """
@@ -44,8 +43,7 @@ See also [`densratio`](@ref).
 Only some estimators define a ratio function that can
 be evaluated outside `x_de`.
 """
-densratiofunc(x_nu, x_de, dre::DensityRatioEstimator;
-              optlib=default_optlib(typeof(dre))) =
+densratiofunc(x_nu, x_de, dre::DensityRatioEstimator; optlib=default_optlib(typeof(dre))) =
   _densratiofunc(x_nu, x_de, dre, optlib)
 
 """
@@ -55,32 +53,24 @@ Return default optimization library for density ratio
 estimator `dre`. The function can also be called on the
 type `typeof(dre)`.
 """
-default_optlib(dre::DensityRatioEstimator) =
-  default_optlib(typeof(dre))
+default_optlib(dre::DensityRatioEstimator) = default_optlib(typeof(dre))
 
 """
     available_optlib(dre)
 
 Return list of implementations available via different
 optimization frameworks.
 """
-available_optlib(dre::DensityRatioEstimator) =
-  available_optlib(typeof(dre))
+available_optlib(dre::DensityRatioEstimator) = available_optlib(typeof(dre))
 
 ###################################################
 ## functions to be implemented by new estimators ##
 ###################################################
 
-_densratio(x_nu, x_de, dre::DensityRatioEstimator,
-           optlib::Type{OptimizationLibrary}) =
-  @error "not implemented"
+_densratio(x_nu, x_de, dre::DensityRatioEstimator, optlib::Type{OptimizationLibrary}) = @error "not implemented"
 
-_densratiofunc(x_nu, x_de, dre::DensityRatioEstimator,
-               optlib::Type{OptimizationLibrary}) =
-  @error "not implemented"
+_densratiofunc(x_nu, x_de, dre::DensityRatioEstimator, optlib::Type{OptimizationLibrary}) = @error "not implemented"
 
-default_optlib(dre::Type{DensityRatioEstimator}) =
-  @error "not implemented"
+default_optlib(dre::Type{DensityRatioEstimator}) = @error "not implemented"
 
-available_optlib(dre::Type{DensityRatioEstimator}) =
-  @error "not implemented"
+available_optlib(dre::Type{DensityRatioEstimator}) = @error "not implemented"

src/api/fitters.jl

@@ -24,14 +24,12 @@ Hyperparameter tuning is not defined for all
 density ratio estimators. Therefore, this
 function may not work with some estimators.
 """
-fit(dre::Type{<:DensityRatioEstimator}, x_nu, x_de,
-    fitter::EstimatorFitter; optlib=default_optlib(dre)) =
+fit(dre::Type{<:DensityRatioEstimator}, x_nu, x_de, fitter::EstimatorFitter; optlib=default_optlib(dre)) =
   _fit(dre, x_nu, x_de, fitter, optlib)
 
 ################################################
 ## functions to be implemented by new fitters ##
 ################################################
 
-_fit(dre::Type{DensityRatioEstimator}, x_nu, x_de,
-     fitter::EstimatorFitter, optlib::Type{OptimizationLibrary}) =
+_fit(dre::Type{DensityRatioEstimator}, x_nu, x_de, fitter::EstimatorFitter, optlib::Type{OptimizationLibrary}) =
   @error "not implemented"

src/api/optlibs.jl

@@ -10,7 +10,7 @@ An optimization library (e.g. Optim.jl, Convex.jl, JuMP.jl).
 abstract type OptimizationLibrary end
 
 # optimization libraries for dispatch
-struct JuliaLib  <: OptimizationLibrary end
-struct OptimLib  <: OptimizationLibrary end
+struct JuliaLib <: OptimizationLibrary end
+struct OptimLib <: OptimizationLibrary end
 struct ConvexLib <: OptimizationLibrary end
-struct JuMPLib   <: OptimizationLibrary end
+struct JuMPLib <: OptimizationLibrary end

src/kliep.jl

@@ -22,24 +22,22 @@ Kullback-Leibler importance estimation procedure (KLIEP).
 * Júlio Hoffimann ([email protected])
 """
 @with_kw struct KLIEP{T,RNG} <: DensityRatioEstimator
-  σ::T=2.0
-  b::Int=10
-  rng::RNG=Random.GLOBAL_RNG
+  σ::T = 2.0
+  b::Int = 10
+  rng::RNG = Random.GLOBAL_RNG
 end
 
 default_optlib(dre::Type{<:KLIEP}) = OptimLib
 
 available_optlib(dre::Type{<:KLIEP}) = [OptimLib, ConvexLib]
 
-function _densratio(x_nu, x_de, dre::KLIEP,
-                    optlib::Type{<:OptimizationLibrary})
+function _densratio(x_nu, x_de, dre::KLIEP, optlib::Type{<:OptimizationLibrary})
   K_nu, K_de, x_ba = _kliep_consts(x_nu, x_de, dre)
   α = _kliep_coeffs(K_nu, K_de, dre, optlib)
-  K_de*α
+  K_de * α
 end
 
-function _densratiofunc(x_nu, x_de, dre::KLIEP,
-                        optlib::Type{<:OptimizationLibrary})
+function _densratiofunc(x_nu, x_de, dre::KLIEP, optlib::Type{<:OptimizationLibrary})
   K_nu, K_de, x_ba = _kliep_consts(x_nu, x_de, dre)
   α = _kliep_coeffs(K_nu, K_de, dre, optlib)
   function r(x)
@@ -62,6 +60,4 @@ end
 
 Return the coefficients of KLIEP basis expansion.
 """
-_kliep_coeffs(K_nu, K_de, dre::KLIEP,
-              optlib::Type{<:OptimizationLibrary}) =
-  @error "not implemented"
+_kliep_coeffs(K_nu, K_de, dre::KLIEP, optlib::Type{<:OptimizationLibrary}) = @error "not implemented"

src/kliep/convex.jl

@@ -17,9 +17,10 @@ function _kliep_coeffs(K_nu, K_de, dre::KLIEP, optlib::Type{ConvexLib})
   k = sum(K_de, dims=1)
 
   # objective function and constraints
-  α = Convex.Variable(b); w = K*α
-  objective   = sum(log(w[i]) for i in 1:n_nu)
-  constraints = [α ≥ 0, dot(α,k) == n_de]
+  α = Convex.Variable(b)
+  w = K * α
+  objective = sum(log(w[i]) for i in 1:n_nu)
+  constraints = [α ≥ 0, dot(α, k) == n_de]
 
   # optimization problem
   problem = Convex.maximize(objective, constraints)

src/kliep/optim.jl

@@ -19,37 +19,37 @@ function _kliep_coeffs(K_nu, K_de, dre::KLIEP, optlib::Type{OptimLib})
   lc = uc = [n_de]
 
   # constants for inequality constraints
-  T  = eltype(K_de)
+  T = eltype(K_de)
   lx = fill(zero(T), b)
   ux = fill(Inf, b)
 
   # objective
-  f(α) = -sum(log, K*α)
+  f(α) = -sum(log, K * α)
   function ∇f!(g, α)
-    p = K*α
+    p = K * α
     for l in 1:b
-      g[l] = -sum(K[j,l] / p[j] for j in 1:n_nu)
+      g[l] = -sum(K[j, l] / p[j] for j in 1:n_nu)
     end
   end
   function ∇²f!(h, α)
-    p = K*α
+    p = K * α
     for k in 1:b, l in 1:b
-      h[k,l] = sum(view(K,:,k) .* view(K,:,l) ./ p)
+      h[k, l] = sum(view(K, :, k) .* view(K, :, l) ./ p)
     end
   end
 
   # equality constraint
-  c!(c, α)    = c  .= A*α
-  J!(J, α)    = J  .= A
+  c!(c, α) = c .= A * α
+  J!(J, α) = J .= A
   H!(H, α, λ) = H .+= 0
 
   # initial guess
-  αₒ = fill(n_de/sum(A), b)
+  αₒ = fill(n_de / sum(A), b)
 
   # optimization problem
-  objective   = TwiceDifferentiable(f, ∇f!, ∇²f!, αₒ)
+  objective = TwiceDifferentiable(f, ∇f!, ∇²f!, αₒ)
   constraints = TwiceDifferentiableConstraints(c!, J!, H!, lx, ux, lc, uc)
-  initguess   = αₒ
+  initguess = αₒ
 
   # solve problem with interior-point primal-dual Newton
   solution = optimize(objective, constraints, initguess, IPNewton())

src/kmm.jl

@@ -20,8 +20,7 @@ function _kmm_consts(x_nu, x_de, dre::AbstractKMM)
   Kdede, typeof(σ)(n_de / n_nu) * sum(Kdenu, dims=2)
 end
 
-function _densratio(x_nu, x_de, dre::AbstractKMM,
-                    optlib::Type{<:OptimizationLibrary})
+function _densratio(x_nu, x_de, dre::AbstractKMM, optlib::Type{<:OptimizationLibrary})
   K, κ = _kmm_consts(x_nu, x_de, dre)
   _kmm_ratios(K, κ, dre, optlib)
 end
@@ -47,8 +46,8 @@ Unconstrained Kernel Mean Matching (KMM).
 * Kai Xu ([email protected])
 """
 @with_kw struct uKMM{T} <: AbstractKMM
-  σ::T=2.0
-  λ::T=0.001
+  σ::T = 2.0
+  λ::T = 0.001
 end
 
 default_optlib(dre::Type{<:uKMM}) = JuliaLib
@@ -78,10 +77,10 @@ Kernel Mean Matching (KMM).
 * Kai Xu ([email protected])
 """
 @with_kw struct KMM{T} <: AbstractKMM
-  σ::T=2.0
-  B::T=Inf
-  ϵ::T=0.01
-  λ::T=0.001
+  σ::T = 2.0
+  B::T = Inf
+  ϵ::T = 0.01
+  λ::T = 0.001
 end
 
 default_optlib(dre::Type{<:KMM}) = JuMPLib

src/kmm/jump.jl

@@ -12,7 +12,7 @@ function _kmm_jump_model(K, κ, dre::AbstractKMM, optlib::Type{JuMPLib})
   # optimization problem
   model = Model(optimizer_with_attributes(Ipopt.Optimizer, "print_level" => 0, "sb" => "yes"))
   @variable(model, β[1:m])
-  @objective(model, Min, (1/2) * dot(β, K*β - 2κ))
+  @objective(model, Min, (1 / 2) * dot(β, K * β - 2κ))
 
   return model, β
 end
@@ -38,7 +38,7 @@ function _kmm_ratios(K, κ, dre::KMM, optlib::Type{JuMPLib})
   # adding constriants
   @constraint(model, 0 .≤ β)
   isinf(B) || @constraint(model, β .≤ B)
-  @constraint(model, (1-ϵ) ≤ mean(β) ≤ (1+ϵ))
+  @constraint(model, (1 - ϵ) ≤ mean(β) ≤ (1 + ϵ))
 
   # solve the problem
   optimize!(model)

src/lcv.jl

@@ -24,13 +24,11 @@ end
 
 LCV(ranges::NamedTuple) = LCV(ranges, 10)
 
-function _fit(::Type{<:KLIEP}, x_nu, x_de,
-              fitter::EstimatorFitter,
-              optlib::Type{<:OptimizationLibrary})
+function _fit(::Type{<:KLIEP}, x_nu, x_de, fitter::EstimatorFitter, optlib::Type{<:OptimizationLibrary})
   # retrieve parameters
   ranges = fitter.ranges
   nfolds = fitter.nfolds
-  npts   = length(x_nu)
+  npts = length(x_nu)
 
   @assert nfolds ≤ npts "number of folds must be smaller than number of numerator samples"
 
@@ -48,8 +46,8 @@ function _fit(::Type{<:KLIEP}, x_nu, x_de,
     # estimate loss with cross-validation
     Ĵₖ = map(1:nfolds) do k
       # training and hold-out samples
-      train = [ind for i in vcat(1:k-1, k+1:nfolds) for ind in folds[i]]
-      hold  = folds[k]
+      train = [ind for i in vcat(1:(k - 1), (k + 1):nfolds) for ind in folds[i]]
+      hold = folds[k]
 
       # perform estimation with training samples
       r = densratiofunc(x_nu[train], x_de, dre, optlib=optlib)

src/lsif.jl

@@ -23,25 +23,23 @@ Importance Estimation
 * Júlio Hoffimann ([email protected])
 """
 @with_kw struct LSIF{T,RNG} <: DensityRatioEstimator
-  σ::T=2.0
-  b::Int=10
-  λ::T=0.001
-  rng::RNG=Random.GLOBAL_RNG
+  σ::T = 2.0
+  b::Int = 10
+  λ::T = 0.001
+  rng::RNG = Random.GLOBAL_RNG
 end
 
 default_optlib(dre::Type{<:LSIF}) = OptimLib
 
 available_optlib(dre::Type{<:LSIF}) = [OptimLib, JuMPLib]
 
-function _densratio(x_nu, x_de, dre::LSIF,
-                    optlib::Type{<:OptimizationLibrary})
+function _densratio(x_nu, x_de, dre::LSIF, optlib::Type{<:OptimizationLibrary})
   K_de, H, h, x_ba = _lsif_consts(x_nu, x_de, dre)
   α = _lsif_coeffs(H, h, dre, optlib)
-  K_de*α
+  K_de * α
 end
 
-function _densratiofunc(x_nu, x_de, dre::LSIF,
-                        optlib::Type{<:OptimizationLibrary})
+function _densratiofunc(x_nu, x_de, dre::LSIF, optlib::Type{<:OptimizationLibrary})
   K_de, H, h, x_ba = _lsif_consts(x_nu, x_de, dre)
   α = _lsif_coeffs(H, h, dre, optlib)
   function r(x)
@@ -61,7 +59,7 @@ function _lsif_consts(x_nu, x_de, dre)
     φ′ = view(K_de, :, l′)
     for l in 1:l′
       φ = view(K_de, :, l)
-      Φ[l,l′] = mean(φ .* φ′)
+      Φ[l, l′] = mean(φ .* φ′)
     end
   end
   H = Symmetric(Φ)
@@ -75,5 +73,4 @@ end
 
 Return the coefficients of LSIF basis expansion.
 """
-_lsif_coeffs(H, h, dre::LSIF, optlib::Type{<:OptimizationLibrary}) =
-  @error "not implemented"
+_lsif_coeffs(H, h, dre::LSIF, optlib::Type{<:OptimizationLibrary}) = @error "not implemented"

src/lsif/jump.jl

@@ -11,7 +11,7 @@ function _lsif_coeffs(H, h, dre::LSIF, optlib::Type{JuMPLib})
 
   model = Model(optimizer_with_attributes(Ipopt.Optimizer, "print_level" => 0, "sb" => "yes"))
   @variable(model, α[1:b])
-  @objective(model, Min, (1/2) * dot(α, H*α - 2h) + λ * sum(α))
+  @objective(model, Min, (1 / 2) * dot(α, H * α - 2h) + λ * sum(α))
   @constraint(model, α .≥ 0)
 
   # solve the problem