Add tests

Author: BatyLeo

src/StochasticVehicleScheduling/StochasticVehicleScheduling.jl

@@ -116,8 +116,10 @@ $TYPEDSIGNATURES
 Return the anticipative solver: a callable `(scenario::VSPScenario; instance, kwargs...) -> y`
 that solves the 1-scenario stochastic VSP.
 """
-function Utils.generate_anticipative_solver(::StochasticVehicleSchedulingBenchmark)
-    return AnticipativeSolver()
+function Utils.generate_anticipative_solver(
+    ::StochasticVehicleSchedulingBenchmark; model_builder=scip_model
+)
+    return AnticipativeSolver(; model_builder=model_builder)
 end
 
 """
@@ -126,9 +128,9 @@ $TYPEDSIGNATURES
 Return the parametric anticipative solver: a callable `(θ, scenario::VSPScenario; instance, kwargs...) -> y`.
 """
 function Utils.generate_parametric_anticipative_solver(
-    ::StochasticVehicleSchedulingBenchmark
+    ::StochasticVehicleSchedulingBenchmark; model_builder=scip_model
 )
-    return AnticipativeSolver()
+    return AnticipativeSolver(; model_builder=model_builder)
 end
 
 """

src/StochasticVehicleScheduling/solution/algorithms/anticipative_solver.jl

@@ -1,4 +1,5 @@
-@kwdef struct AnticipativeSolver{A}
+@kwdef struct AnticipativeSolver{M,A}
+    model_builder::M = scip_model
     single_scenario_algorithm::A = compact_mip
 end
 
@@ -8,10 +9,14 @@ end
 
 function (solver::AnticipativeSolver)(scenario; instance::Instance, kwargs...)
     stochastic_inst = build_stochastic_instance(instance, [scenario])
-    return solver.single_scenario_algorithm(stochastic_inst)
+    return solver.single_scenario_algorithm(
+        stochastic_inst; model_builder=solver.model_builder, kwargs...
+    )
 end
 
 function (solver::AnticipativeSolver)(θ, scenario; instance::Instance, kwargs...)
     stochastic_inst = build_stochastic_instance(instance, [scenario])
-    return solver.single_scenario_algorithm(stochastic_inst, θ)
+    return solver.single_scenario_algorithm(
+        stochastic_inst, θ; model_builder=solver.model_builder, kwargs...
+    )
 end

test/contextual_stochastic_argmax.jl

@@ -38,6 +38,31 @@
     @test first(d1).x ≈ first(d2).x
 end
 
+@testset "Parametric Anticipative Solver - ContextualStochasticArgmax" begin
+    using DecisionFocusedLearningBenchmarks
+
+    b = ContextualStochasticArgmaxBenchmark(; n=5, d=3, seed=0)
+    dataset = generate_dataset(b, 2; contexts_per_instance=1, nb_scenarios=1)
+    sample = first(dataset)
+    scenario = generate_scenario(b, StableRNG(0); sample.context...)
+
+    solver = generate_anticipative_solver(b)
+    parametric_solver = generate_parametric_anticipative_solver(b)
+
+    # 1. Zero perturbation equivalence
+    θ_zero = zeros(eltype(scenario), size(scenario))
+    @test parametric_solver(θ_zero, scenario; sample.context...) ==
+        solver(scenario; sample.context...)
+
+    # 2. Extreme perturbation
+    θ_extreme = zeros(eltype(scenario), size(scenario))
+    θ_extreme[1] = 1000.0  # Force dimension 1
+    y_extreme = parametric_solver(θ_extreme, scenario; sample.context...)
+
+    @test y_extreme[1] == 1.0     # Only dimension 1 should be active
+    @test sum(y_extreme) ≈ 1.0    # One-hot preserved
+end
+
 @testset "csa_saa_policy" begin
     using DecisionFocusedLearningBenchmarks
 

test/vsp.jl

@@ -95,4 +95,28 @@
     sample = unlabeled[1]
     y_anticipative = anticipative_solver(sample.scenario; sample.context...)
     @test y_anticipative isa BitVector
+
+    # Extract necessary dependencies
+    parametric_solver = generate_parametric_anticipative_solver(b)
+    nb_edges = ne(sample.instance.graph)
+
+    # 1. Zero perturbation equivalence
+    θ_zero = zeros(nb_edges)
+    y_zero = parametric_solver(θ_zero, sample.scenario; sample.context...)
+
+    @test y_zero == y_anticipative
+    @test y_zero isa BitVector
+
+    # 2. Perturbation execution
+    θ_random = randn(nb_edges)
+    y_rand = parametric_solver(θ_random, sample.scenario; sample.context...)
+
+    @test length(y_rand) == nb_edges
+    @test y_rand isa BitVector
+
+    # 3. High negative perturbation on edge forces activation (it's minimization)
+    θ_extreme = zeros(nb_edges)
+    θ_extreme[1] = -100000.0  # large negative pull for edge 1
+    y_extreme = parametric_solver(θ_extreme, sample.scenario; sample.context...)
+    @test y_extreme[1] == 1.0 # BitVector
 end