merge models

ensure a model and the corresponding nonlinear least-squares
formulation represent the same problem and are based on the
same data.

Author: dpo

src/bpdn_model.jl

@@ -16,11 +16,12 @@ bpdn_data(compound::Int = 1, args...) =
   bpdn_data(200 * compound, 512 * compound, 10 * compound, args...)
 
 """
-    model, sol = bpdn_model(args...)
-    model, sol = bpdn_model(compound = 1, args...)
+    model, nls_model, sol = bpdn_model(args...)
+    model, nls_model, sol = bpdn_model(compound = 1, args...)
 
-Return an instance of an `NLPModel` representing the basis-pursuit denoise
-problem, i.e., the under-determined linear least-squares objective
+Return an instance of an `NLPModel` and an instance of an `NLSModel` representing
+the same basis-pursuit denoise problem, i.e., the under-determined linear
+least-squares objective
 
    ½ ‖Ax - b‖₂²,
 
@@ -32,7 +33,7 @@ vector following a normal distribution with mean zero and standard deviation σ.
 * `m :: Int`: the number of rows of A
 * `n :: Int`: the number of columns of A (with `n` ≥ `m`)
 * `k :: Int`: the number of nonzero elements in x̄
-* `noise :: Float64`: noise amount ϵ (default: 0.01).
+* `noise :: Float64`: noise standard deviation σ (default: 0.01).
 
 The second form calls the first form with arguments
 
@@ -42,8 +43,8 @@ The second form calls the first form with arguments
 
 ## Return Value
 
-An instance of a `FirstOrderModel` that represents the basis-pursuit denoise problem
-and the exact solution x̄.
+An instance of a `FirstOrderModel` and of a `FirstOrderNLSModel` that represent the same
+basis-pursuit denoise problem, and the exact solution x̄.
 """
 function bpdn_model(args...)
   A, b, b0, x0 = bpdn_data(args...)
@@ -55,6 +56,9 @@ function bpdn_model(args...)
     r
   end
 
+  jprod_resid!(Jv, x, v) = mul!(Jv, A, v)
+  jtprod_resid!(Jtv, x, v) = mul!(Jtv, A', v)
+
   function obj(x)
     resid!(r, x)
     dot(r, r) / 2
@@ -66,32 +70,6 @@ function bpdn_model(args...)
     g
   end
 
-  FirstOrderModel(obj, grad!, zero(x0), name = "BPDN"), x0
+  FirstOrderModel(obj, grad!, zero(x0), name = "BPDN"), FirstOrderNLSModel(resid!, jprod_resid!, jtprod_resid!, size(A, 1), zero(x0), name = "BPDN-LS"), x0
 end
 
-"""
-    model, sol = bpdn_nls_model(args...)
-    model, sol = bpdn_nls_model(compound = 1, args...)
-
-Return an instance of a `FirstOrderNLSModel` that represents the basis-pursuit
-denoise problem explicitly as a least-squares problem and the exact solution x̄.
-
-See the documentation of `bpdn_model()` for more information and a
-description of the arguments.
-"""
-function bpdn_nls_model(args...)
-  A, b, b0, x0 = bpdn_data(args...)
-  r = similar(b)
-
-  function resid!(r, x)
-    mul!(r, A, x)
-    r .-= b
-    r
-  end
-
-  jprod_resid!(Jv, x, v) = mul!(Jv, A, v)
-  jtprod_resid!(Jtv, x, v) = mul!(Jtv, A', v)
-
-  FirstOrderNLSModel(resid!, jprod_resid!, jtprod_resid!, size(A, 1), zero(x0), name = "BPDN-LS"),
-  x0
-end

src/fh_model.jl

@@ -64,8 +64,9 @@ end
 """
     fh_model(; kwargs...)
 
-Return an instance of an `NLPModel` representing the Fitzhugh-Nagumo problem,
-i.e., the over-determined nonlinear least-squares objective
+Return an instance of an `NLPModel` and an instance of an `NLSModel` representing
+the same Fitzhugh-Nagumo problem, i.e., the over-determined nonlinear
+least-squares objective
 
    ½ ‖F(x)‖₂²,
 
@@ -80,24 +81,12 @@ constructure, e.g., to set the automatic differentiation backend.
 
 ## Return Value
 
-An instance of an `ADNLPModel` that represents the Fitzhugh-Nagumo problem.
+An instance of an `ADNLPModel` that represents the Fitzhugh-Nagumo problem, an instance
+of an `ADNLSModel` that represents the same problem, and the exact solution.
 """
 function fh_model(; kwargs...)
-  data, simulate, resid, misfit, x0 = FH_smooth_term()
-  ADNLPModels.ADNLPModel(misfit, ones(5); kwargs...), x0
-end
-
-"""
-    fh_nls_model(; kwargs...)
-
-Return an instance of an `ADNLSModel` that represents the Fitzhugh-Nagumo
-problem explicitly as a nonlinear least-squares problem.
-
-See the documentation of `fh_model()` for more information and a
-description of the arguments.
-"""
-function fh_nls_model(; kwargs...)
   data, simulate, resid, misfit, x0 = FH_smooth_term()
   nequ = 202
-  ADNLPModels.ADNLSModel(resid, ones(5), nequ; kwargs...), x0
+  ADNLPModels.ADNLPModel(misfit, ones(5); kwargs...), ADNLPModels.ADNLSModel(resid, ones(5), nequ; kwargs...), x0
 end
+

test/runtests.jl

@@ -3,39 +3,29 @@ using ADNLPModels, DifferentialEquations
 using RegularizedProblems
 
 @testset "BPDN" begin
-  model, sol = bpdn_model()
+  model, nls_model, sol = bpdn_model()
   @test typeof(model) <: FirstOrderModel
   @test typeof(sol) == Vector{Float64}
   @test model.meta.nvar == 512
   @test all(model.meta.x0 .== 0)
   @test length(findall(x -> x .!= 0, sol)) == 10
-end
 
-@testset "BPDN-LS" begin
-  model, sol = bpdn_nls_model()
-  @test typeof(model) <: FirstOrderNLSModel
-  @test typeof(sol) == Vector{Float64}
-  @test model.meta.nvar == 512
-  @test model.nls_meta.nequ == 200
-  @test all(model.meta.x0 .== 0)
-  @test length(findall(x -> x .!= 0, sol)) == 10
+  @test typeof(nls_model) <: FirstOrderNLSModel
+  @test nls_model.meta.nvar == 512
+  @test nls_model.nls_meta.nequ == 200
+  @test all(nls_model.meta.x0 .== 0)
 end
 
 @testset "FH" begin
-  model, sol = fh_model()
+  model, nls_model, sol = fh_model()
   @test typeof(model) <: ADNLPModel
   @test typeof(sol) == Vector{Float64}
   @test model.meta.nvar == 5
   @test all(model.meta.x0 .== 1)
   @test length(findall(x -> x .!= 0, sol)) == 2
-end
 
-@testset "FH-NLS" begin
-  model, sol = fh_nls_model()
-  @test typeof(model) <: ADNLSModel
-  @test typeof(sol) == Vector{Float64}
-  @test model.meta.nvar == 5
-  @test model.nls_meta.nequ == 202
-  @test all(model.meta.x0 .== 1)
-  @test length(findall(x -> x .!= 0, sol)) == 2
+  @test typeof(nls_model) <: ADNLSModel
+  @test nls_model.meta.nvar == 5
+  @test nls_model.nls_meta.nequ == 202
+  @test all(nls_model.meta.x0 .== 1)
 end