[WIP] Refactor Tests

test/runtests.jl

@@ -4,6 +4,7 @@ using ProximalOperators
 using ADNLPModels,
   OptimizationProblems,
   OptimizationProblems.ADNLPProblems,
+  ManualNLPModels,
   NLPModels,
   NLPModelsModifiers,
   RegularizedProblems,
@@ -18,6 +19,10 @@ const global bpdn, bpdn_nls, sol = bpdn_model(compound)
 const global bpdn2, bpdn_nls2, sol2 = bpdn_model(compound, bounds = true)
 const global λ = norm(grad(bpdn, zeros(bpdn.meta.nvar)), Inf) / 10
 
+include("utils.jl")
+include("test-solver.jl")
+
+include("test-R2N.jl")
 include("test_AL.jl")
 
 for (mod, mod_name) ∈ ((x -> x, "exact"), (LSR1Model, "lsr1"), (LBFGSModel, "lbfgs"))

test/test-R2N.jl

@@ -0,0 +1,83 @@
+@testset "R2N" begin
+  # BASIC TESTS
+  # Test basic NLP with 2-norm
+  @testset "BASIC" begin
+    rosenbrock_nlp = construct_rosenbrock_nlp()
+    rosenbrock_reg_nlp = RegularizedNLPModel(rosenbrock_nlp, NormL2(0.01))
+
+    # Test first order status
+    first_order_kwargs = (atol = 1e-6, rtol = 1e-6)
+    test_solver(
+      rosenbrock_reg_nlp,
+      "R2N",
+      expected_status = :first_order,
+      solver_kwargs = first_order_kwargs,
+    )
+    solver, stats = R2NSolver(rosenbrock_reg_nlp), RegularizedExecutionStats(rosenbrock_reg_nlp)
+
+    # Test max time status
+    max_time_kwargs = (x0 = [π, -π], atol = 1e-16, rtol = 1e-16, max_time = 1e-12)
+    test_solver(
+      rosenbrock_reg_nlp,
+      "R2N",
+      expected_status = :max_time,
+      solver_kwargs = max_time_kwargs,
+    )
+
+    # Test max iter status
+    max_iter_kwargs = (x0 = [π, -π], atol = 1e-16, rtol = 1e-16, max_iter = 1)
+    test_solver(
+      rosenbrock_reg_nlp,
+      "R2N",
+      expected_status = :max_iter,
+      solver_kwargs = max_iter_kwargs,
+    )
+
+    # Test max eval status
+    max_eval_kwargs = (x0 = [π, -π], atol = 1e-16, rtol = 1e-16, max_eval = 1)
+    test_solver(
+      rosenbrock_reg_nlp,
+      "R2N",
+      expected_status = :max_eval,
+      solver_kwargs = max_eval_kwargs,
+    )
+  end
+  # BPDN TESTS
+
+  # Test bpdn with L-BFGS and 1-norm
+  @testset "BPDN" begin
+    bpdn_kwargs = (x0 = zeros(bpdn.meta.nvar), σk = 1.0, β = 1e16, atol = 1e-6, rtol = 1e-6)
+    reg_nlp = RegularizedNLPModel(LBFGSModel(bpdn), NormL1(λ))
+    test_solver(reg_nlp, "R2N", expected_status = :first_order, solver_kwargs = bpdn_kwargs)
+    solver, stats = R2NSolver(reg_nlp), RegularizedExecutionStats(reg_nlp)
+    @test @wrappedallocs(
+      solve!(solver, reg_nlp, stats, σk = 1.0, β = 1e16, atol = 1e-6, rtol = 1e-6)
+    ) == 0
+
+    #test_solver(reg_nlp,  # FIXME
+    #            "R2N", 
+    #            expected_status = :first_order,
+    #            solver_kwargs=bpdn_kwargs,
+    #            solver_constructor_kwargs=(subsolver=R2DHSolver,))
+
+    # Test bpdn with L-SR1 and 0-norm
+    reg_nlp = RegularizedNLPModel(LSR1Model(bpdn), NormL0(λ))
+    test_solver(reg_nlp, "R2N", expected_status = :first_order, solver_kwargs = bpdn_kwargs)
+    solver, stats = R2NSolver(reg_nlp), RegularizedExecutionStats(reg_nlp)
+    @test @wrappedallocs(
+      solve!(solver, reg_nlp, stats, σk = 1.0, β = 1e16, atol = 1e-6, rtol = 1e-6)
+    ) == 0
+
+    test_solver(
+      reg_nlp,
+      "R2N",
+      expected_status = :first_order,
+      solver_kwargs = bpdn_kwargs,
+      solver_constructor_kwargs = (subsolver = R2DHSolver,),
+    )
+    solver, stats = R2NSolver(reg_nlp, subsolver = R2DHSolver), RegularizedExecutionStats(reg_nlp)
+    @test @wrappedallocs(
+      solve!(solver, reg_nlp, stats, σk = 1.0, β = 1e16, atol = 1e-6, rtol = 1e-6)
+    ) == 0
+  end
+end

test/test-solver.jl

@@ -0,0 +1,47 @@
+function test_solver(
+  reg_nlp::R,
+  solver_name::String;
+  expected_status = :first_order,
+  solver_constructor_kwargs = (;),
+  solver_kwargs = (;),
+) where {R}
+
+  # Test output with allocating calling form
+  solver_fun = getfield(RegularizedOptimization, Symbol(solver_name))
+  stats_basic = solver_fun(
+    reg_nlp.model,
+    reg_nlp.h,
+    ROSolverOptions();
+    solver_constructor_kwargs...,
+    solver_kwargs...,
+  )
+
+  x0 = get(solver_kwargs, :x0, reg_nlp.model.meta.x0)
+  @test typeof(stats_basic.solution) == typeof(x0)
+  @test length(stats_basic.solution) == reg_nlp.model.meta.nvar
+  @test typeof(stats_basic.dual_feas) == eltype(stats_basic.solution)
+  @test stats_basic.status == expected_status
+  @test obj(reg_nlp, stats_basic.solution) == stats_basic.objective
+  @test stats_basic.objective <= obj(reg_nlp, x0)
+
+  # Test output with optimized calling form
+  solver_constructor = getfield(RegularizedOptimization, Symbol(solver_name * "Solver"))
+  solver = solver_constructor(reg_nlp; solver_constructor_kwargs...)
+  stats_optimized = RegularizedExecutionStats(reg_nlp)
+
+  # Remove the x0 entry from solver_kwargs
+  optimized_solver_kwargs = Base.structdiff(solver_kwargs, NamedTuple{(:x0,)})
+  solve!(solver, reg_nlp, stats_optimized; x = x0, optimized_solver_kwargs...) # It would be interesting to check for allocations here as well but depending on 
+  # the structure of solver_kwargs, some variables might get boxed, resulting in 
+  # false positives, for example if tol = 1e-3; solver_kwargs = (atol = tol),
+  # then wrappedallocs would give a > 0 answer...
+  @test typeof(stats_optimized.solution) == typeof(x0)
+  @test length(stats_optimized.solution) == reg_nlp.model.meta.nvar
+  @test typeof(stats_optimized.dual_feas) == eltype(stats_optimized.solution)
+  @test stats_optimized.status == expected_status
+  @test obj(reg_nlp, stats_optimized.solution) == stats_optimized.objective
+  @test stats_optimized.objective <= obj(reg_nlp, x0)
+
+  # TODO: test that the optimized entries in stats_optimized and stats_basic are the same.
+
+end

test/test_allocs.jl

@@ -1,44 +1,3 @@
-"""
-    @wrappedallocs(expr)
-
-Given an expression, this macro wraps that expression inside a new function
-which will evaluate that expression and measure the amount of memory allocated
-by the expression. Wrapping the expression in a new function allows for more
-accurate memory allocation detection when using global variables (e.g. when
-at the REPL).
-
-This code is based on that of https://github.com/JuliaAlgebra/TypedPolynomials.jl/blob/master/test/runtests.jl
-
-For example, `@wrappedallocs(x + y)` produces:
-
-```julia
-function g(x1, x2)
-    @allocated x1 + x2
-end
-g(x, y)
-```
-
-You can use this macro in a unit test to verify that a function does not
-allocate:
-
-```
-@test @wrappedallocs(x + y) == 0
-```
-"""
-macro wrappedallocs(expr)
-  kwargs = [a for a in expr.args if isa(a, Expr)]
-  args = [a for a in expr.args if isa(a, Symbol)]
-
-  argnames = [gensym() for a in args]
-  kwargs_dict = Dict{Symbol, Any}(a.args[1] => a.args[2] for a in kwargs if a.head == :kw)
-  quote
-    function g($(argnames...); kwargs_dict...)
-      $(Expr(expr.head, argnames..., kwargs...)) # Call the function twice to make the allocated macro more stable
-      @allocated $(Expr(expr.head, argnames..., kwargs...))
-    end
-    $(Expr(:call, :g, [esc(a) for a in args]...))
-  end
-end
 
 # Test non allocating solve!
 @testset "NLP allocs" begin

test/utils.jl

@@ -0,0 +1,52 @@
+"""
+    @wrappedallocs(expr)
+
+Given an expression, this macro wraps that expression inside a new function
+which will evaluate that expression and measure the amount of memory allocated
+by the expression. Wrapping the expression in a new function allows for more
+accurate memory allocation detection when using global variables (e.g. when
+at the REPL).
+
+This code is based on that of https://github.com/JuliaAlgebra/TypedPolynomials.jl/blob/master/test/runtests.jl
+
+You can use this macro in a unit test to verify that a function does not
+allocate:
+
+```
+@test @wrappedallocs(x + y) == 0
+```
+"""
+macro wrappedallocs(expr)
+  kwargs = [a for a in expr.args if isa(a, Expr)]
+  args = [a for a in expr.args if isa(a, Symbol)]
+
+  argnames = [gensym() for a in args]
+  kwargs_dict = Dict{Symbol, Any}(a.args[1] => a.args[2] for a in kwargs if a.head == :kw)
+  quote
+    function g($(argnames...); kwargs_dict...)
+      $(Expr(expr.head, argnames..., kwargs...)) # Call the function twice to make the allocated macro more stable
+      @allocated $(Expr(expr.head, argnames..., kwargs...))
+    end
+    $(Expr(:call, :g, [esc(a) for a in args]...))
+  end
+end
+
+# Construct the rosenbrock problem.
+
+function rosenbrock_f(x::Vector{T}) where {T <: Real}
+  100 * (x[2] - x[1]^2)^2 + (1 - x[1])^2
+end
+
+function rosenbrock_grad!(gx::Vector{T}, x::Vector{T}) where {T <: Real}
+  gx[1] = -400 * x[1] * (x[2] - x[1]^2) - 2 * (1 - x[1])
+  gx[2] = 200 * (x[2] - x[1]^2)
+end
+
+function rosenbrock_hv!(hv::Vector{T}, x::Vector{T}, v::Vector{T}; obj_weight = 1.0) where {T}
+  hv[1] = (1200 * x[1]^2 - 400 * x[2] + 2) * v[1] - 400 * x[1] * v[2]
+  hv[2] = -400 * x[1] * v[1] + 200 * v[2]
+end
+
+function construct_rosenbrock_nlp()
+  return NLPModel(zeros(2), rosenbrock_f, grad = rosenbrock_grad!, hprod = rosenbrock_hv!)
+end