Merge pull request #361 from oscardssmith/fix-TrustRegion-autodiff

fix `autodiff=AutoFiniteDiff()` for `TrustRegion`

Author: avik-pal

src/algorithms/trust_region.jl

@@ -24,9 +24,16 @@ function TrustRegion(; concrete_jac = nothing, linsolve = nothing, precs = DEFAU
         initial_trust_radius::Real = 0 // 1, step_threshold::Real = 1 // 10000,
         shrink_threshold::Real = 1 // 4, expand_threshold::Real = 3 // 4,
         shrink_factor::Real = 1 // 4, expand_factor::Real = 2 // 1,
-        max_shrink_times::Int = 32, vjp_autodiff = nothing, autodiff = nothing)
+        max_shrink_times::Int = 32, autodiff = nothing, vjp_autodiff = nothing)
     descent = Dogleg(; linsolve, precs)
-    forward_ad = autodiff isa ADTypes.AbstractForwardMode ? autodiff : nothing
+    if autodiff isa Union{ADTypes.AbstractForwardMode, ADTypes.AbstractFiniteDifferencesMode}
+        forward_ad = autodiff
+    else
+        forward_ad = nothing
+    end
+    if isnothing(vjp_autodiff) && autodiff isa ADTypes.AbstractFiniteDifferencesMode
+        vjp_autodiff = autodiff
+    end
     trustregion = GenericTrustRegionScheme(; method = radius_update_scheme, step_threshold,
         shrink_threshold, expand_threshold, shrink_factor, expand_factor,
         reverse_ad = vjp_autodiff, forward_ad)

test/misc/polyalgs.jl

@@ -93,11 +93,14 @@ end
         maxiters = 10)
 end
 
+no_ad_fast = FastShortcutNonlinearPolyalg(autodiff=AutoFiniteDiff())
+no_ad_robust = RobustMultiNewton(autodiff=AutoFiniteDiff())
+no_ad_algs = Set([no_ad_fast, no_ad_robust, no_ad_fast.algs..., no_ad_robust.algs...])
 @testset "[IIP] no AD" begin
     f_iip = Base.Experimental.@opaque (du, u, p) -> du .= u .* u .- p
-    u0 = [0.0]
+    u0 = [0.5]
     prob = NonlinearProblem(f_iip, u0, 1.0)
-    for alg in [RobustMultiNewton(autodiff = AutoFiniteDiff())]
+    for alg in no_ad_algs
         sol = solve(prob, alg)
         @test isapprox(only(sol.u), 1.0)
         @test SciMLBase.successful_retcode(sol.retcode)
@@ -106,9 +109,9 @@ end
 
 @testset "[OOP] no AD" begin
     f_oop = Base.Experimental.@opaque (u, p) -> u .* u .- p
-    u0 = [0.0]
+    u0 = [0.5]
     prob = NonlinearProblem{false}(f_oop, u0, 1.0)
-    for alg in [RobustMultiNewton(autodiff = AutoFiniteDiff())]
+    for alg in no_ad_algs
         sol = solve(prob, alg)
         @test isapprox(only(sol.u), 1.0)
         @test SciMLBase.successful_retcode(sol.retcode)