Fix optimizer factory usage in Sandwiching (#148)

Author: odow

ext/MultiObjectiveAlgorithmsPolyhedraExt.jl

@@ -45,9 +45,10 @@ function MOA.minimize_multiobjective!(
     n = MOI.output_dimension(model.f)
     scalars = MOI.Utilities.scalarize(model.f)
     status = MOI.OPTIMAL
-    optimizer = typeof(model.inner.optimizer)
-    δ_OPS_optimizer = optimizer()
-    MOI.set(δ_OPS_optimizer, MOI.Silent(), true)
+    δ_OPS_optimizer = MOI.instantiate(model.optimizer_factory)
+    if MOI.supports(δ_OPS_optimizer, MOI.Silent())
+        MOI.set(δ_OPS_optimizer, MOI.Silent(), true)
+    end
     y = MOI.add_variables(δ_OPS_optimizer, n)
     anchors = Dict{Vector{Float64},Dict{MOI.VariableIndex,Float64}}()
     yI, yUB = zeros(n), zeros(n)

src/MultiObjectiveAlgorithms.jl

@@ -124,6 +124,7 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
     ideal_point::Vector{Float64}
     compute_ideal_point::Bool
     subproblem_count::Int
+    optimizer_factory::Any
 
     function Optimizer(optimizer_factory)
         return new(
@@ -137,6 +138,7 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
             Float64[],
             default(ComputeIdealPoint()),
             0,
+            optimizer_factory,
         )
     end
 end

src/algorithms/Sandwiching.jl

@@ -6,7 +6,10 @@
 """
     Sandwiching(precision::Float64)
 
-An algorithm that implemennts the paper described in Koenen, M., Balvert, M., & Fleuren, H. A. (2023). A Renewed Take on Weighted Sum in Sandwich Algorithms: Modification of the Criterion Space. (Center Discussion Paper; Vol. 2023-012). CentER, Center for Economic Research.
+An algorithm that implemennts the paper described in Koenen, M., Balvert, M., &
+Fleuren, H. A. (2023). A Renewed Take on Weighted Sum in Sandwich Algorithms:
+Modification of the Criterion Space. (Center Discussion Paper; Vol. 2023-012).
+CentER, Center for Economic Research.
 
 ## Compat
 

test/algorithms/Sandwiching.jl

@@ -12,6 +12,8 @@ import MultiObjectiveAlgorithms as MOA
 import MultiObjectiveAlgorithms: MOI
 import Polyhedra
 
+include(joinpath(dirname(@__DIR__), "mock_optimizer.jl"))
+
 function run_tests()
     for name in names(@__MODULE__; all = true)
         if startswith("$name", "test_")
@@ -190,6 +192,35 @@ function test_time_limit()
     return
 end
 
+function test_solve_failures()
+    m, n = 2, 10
+    p1 = [5.0 1 10 8 3 5 3 3 7 2; 10 6 1 6 8 3 2 10 6 1]
+    p2 = [4.0 6 4 3 1 6 8 2 9 7; 8 8 8 2 4 8 8 1 10 1]
+    w = [5.0 9 3 5 10 5 7 10 7 8; 4 8 8 6 10 8 10 7 5 1]
+    b = [34.0, 33.0]
+    for fail_after in 0:4
+        model = MOA.Optimizer(mock_optimizer(fail_after))
+        MOI.set(model, MOA.Algorithm(), MOA.Sandwiching(0.0))
+        x_ = MOI.add_variables(model, m * n)
+        x = reshape(x_, m, n)
+        MOI.add_constraint.(model, x, MOI.Interval(0.0, 1.0))
+        f = MOI.Utilities.operate(vcat, Float64, sum(p1 .* x), sum(p2 .* x))
+        MOI.set(model, MOI.ObjectiveSense(), MOI.MAX_SENSE)
+        MOI.set(model, MOI.ObjectiveFunction{typeof(f)}(), f)
+        for i in 1:m
+            f_i = sum(w[i, j] * x[i, j] for j in 1:n)
+            MOI.add_constraint(model, f_i, MOI.LessThan(b[i]))
+        end
+        for j in 1:n
+            MOI.add_constraint(model, sum(1.0 .* x[:, j]), MOI.EqualTo(1.0))
+        end
+        MOI.optimize!(model)
+        @test MOI.get(model, MOI.TerminationStatus()) == MOI.NUMERICAL_ERROR
+        @test MOI.get(model, MOI.ResultCount()) == 0
+    end
+    return
+end
+
 end  # TestSandwiching
 
 TestSandwiching.run_tests()