fix: implement DiffEqBase.anyeltypedual for MTKParameters

Author: AayushSabharwal

docs/src/examples/remake.md

@@ -51,11 +51,8 @@ function loss(x, p)
     odeprob = p[1] # ODEProblem stored as parameters to avoid using global variables
     ps = parameter_values(odeprob) # obtain the parameter object from the problem
     ps = replace(Tunable(), ps, x) # create a copy with the values passed to the loss function
-    T = eltype(x)
-    # we also have to convert the `u0` vector
-    u0 = T.(state_values(odeprob))
     # remake the problem, passing in our new parameter object
-    newprob = remake(odeprob; u0 = u0, p = ps)
+    newprob = remake(odeprob; p = ps)
     timesteps = p[2]
     sol = solve(newprob, AutoTsit5(Rosenbrock23()); saveat = timesteps)
     truth = p[3]

src/systems/parameter_buffer.jl

@@ -384,6 +384,15 @@ function SymbolicIndexingInterface.remake_buffer(sys, oldbuf::MTKParameters, val
     return newbuf
 end
 
+function DiffEqBase.anyeltypedual(
+        p::MTKParameters, ::Type{Val{counter}} = Val{0}) where {counter}
+    DiffEqBase.anyeltypedual(p.tunable)
+end
+function DiffEqBase.anyeltypedual(p::Type{<:MTKParameters{T}},
+        ::Type{Val{counter}} = Val{0}) where {counter} where {T}
+    DiffEqBase.__anyeltypedual(T)
+end
+
 _subarrays(v::AbstractVector) = isempty(v) ? () : (v,)
 _subarrays(v::ArrayPartition) = v.x
 _subarrays(v::Tuple) = v

test/mtkparameters.jl

@@ -2,6 +2,7 @@ using ModelingToolkit
 using ModelingToolkit: t_nounits as t, D_nounits as D, MTKParameters
 using SymbolicIndexingInterface
 using SciMLStructures: SciMLStructures, canonicalize, Tunable, Discrete, Constants
+using OrdinaryDiffEq
 using ForwardDiff
 
 @parameters a b c d::Integer e[1:3] f[1:3, 1:3]::Int g::Vector{AbstractFloat} h::String
@@ -134,3 +135,24 @@ ps = [p => 1.0] # Value for `d` is missing
 
 @test_throws ModelingToolkit.MissingVariablesError ODEProblem(sys, u0, tspan, ps)
 @test_nowarn ODEProblem(sys, u0, tspan, [ps..., d => 1.0])
+
+# Issue#2642
+@parameters α β γ δ
+@variables x(t) y(t)
+eqs = [D(x) ~ (α - β * y) * x
+       D(y) ~ (δ * x - γ) * y]
+@mtkbuild odesys = ODESystem(eqs, t)
+odeprob = ODEProblem(
+    odesys, [x => 1.0, y => 1.0], (0.0, 10.0), [α => 1.5, β => 1.0, γ => 3.0, δ => 1.0])
+tunables, _... = canonicalize(Tunable(), odeprob.p)
+@test tunables isa AbstractVector{Float64}
+
+function loss(x)
+    ps = odeprob.p
+    newps = SciMLStructures.replace(Tunable(), ps, x)
+    newprob = remake(odeprob, p = newps)
+    sol = solve(newprob, Tsit5())
+    return sum(sol)
+end
+
+@test_nowarn ForwardDiff.gradient(loss, collect(tunables))