Merge pull request #18 from fernandopalafox/feature/jacobian_error_ha…

…ndling

Add error handling for missing Jacobian

src/autodiff.jl

@@ -2,6 +2,8 @@ function _solve_jacobian_θ(problem, solution, θ; active_tolerance = 1e-3)
     (; jacobian_z!, jacobian_θ!, lower_bounds, upper_bounds) = problem
     z_star = solution.z
 
+    !isnothing(jacobian_θ!) || throw(ArgumentError("Missing sensitivities. Set `compute_sensitivities = true` when constructing the ParametricMCP."))
+
     inactive_indices = let
         lower_inactive = z_star .>= (lower_bounds .+ active_tolerance)
         upper_inactive = z_star .<= (upper_bounds .- active_tolerance)

test/runtests.jl

@@ -1,5 +1,5 @@
 using ParametricMCPs
-using Test: @testset, @test
+using Test: @testset, @test, @test_throws
 using Random: Random
 using LinearAlgebra: norm
 using Zygote: Zygote
@@ -15,6 +15,7 @@ using FiniteDiff: FiniteDiff
     lower_bounds = [-Inf, -Inf, 0, 0]
     upper_bounds = [Inf, Inf, Inf, Inf]
     problem = ParametricMCPs.ParametricMCP(f, lower_bounds, upper_bounds, parameter_dimension)
+    problem_no_jacobian = ParametricMCPs.ParametricMCP(f, lower_bounds, upper_bounds, parameter_dimension; compute_sensitivities = false)
 
     feasible_parameters = [[0.0, 0.0], [1.0, 0.0], [0.0, 1.0], [rand(rng, 2) for _ in 1:10]...]
     infeasible_parameters = -feasible_parameters
@@ -44,5 +45,14 @@ using FiniteDiff: FiniteDiff
             @test isapprox(∇_autodiff_reverse, ∇_finitediff; atol = 1e-4)
             @test isapprox(∇_autodiff_reverse, ∇_autodiff_forward; atol = 1e-4)
         end
-    end
+    end   
+
+    @testset "missing jacobian" begin
+        function dummy_pipeline(θ, problem)
+            solution = ParametricMCPs.solve(problem, θ)
+            sum(solution.z .^ 2)
+        end
+
+        @test_throws ArgumentError Zygote.gradient(θ -> dummy_pipeline(θ, problem_no_jacobian), feasible_parameters[1])
+    end 
 end