Arno Strouwen

Adding safetestsets showed some potentially unwanted interaction between tests. `f(t, y_nv, ydot_nv, user_data)` is not really being used, instead it seems like a previously defined 3 argument version is doing...

Lux branching

6

```jl using Lux, ComponentArrays, OrdinaryDiffEq, SciMLSensitivity, Random#, Statistics rng = Random.default_rng() tspan = (0.0f0, 8.0f0) ann = Chain(Dense(1, 32, tanh), Dense(32, 32, tanh), Dense(32, 1)) ps, st = Lux.setup(rng, ann)...

use deepcopy and Enzyme.make_zero in automatic default alg

2

Made these changes on my Windows machine, so unable to test locally.

EnzymeVJP often fails is automatic sensealg determination

2

Deleting the sensealg keyword in e.g. this tutorial: https://docs.sciml.ai/SciMLSensitivity/dev/examples/optimal_control/optimal_control/ gives the warning that `EnzymeVJP` does not work on the problem. However, manually specifying Enzyme, e.g. `InterpolatingAdjoint(autojacvec = EnzymeVJP)`, seems to...

```jl using SciMLSensitivity using OrdinaryDiffEq using Zygote dynamics = (x, p, t) -> -x*p[1] function loss1(params) u0 = [1.0+1.0im] problem = ODEProblem(dynamics, u0, (0.0, 1.0), params) rollout = solve(problem, Tsit5(),...

```jl using SciMLSensitivity using OrdinaryDiffEq using Zygote dynamics = (x, _p, _t) -> x function loss1(params) u0 = zeros(2) problem = ODEProblem(dynamics, u0, (0.0, 1.0), params) rollout = solve(problem, Tsit5(),...

```jl dynamics = (x, _p, _t) -> x function loss1(params) u0 = zeros(2) problem = ODEProblem(dynamics, u0, (0.0, 1.0), params) rollout = solve(problem, Tsit5(), u0 = u0, p = params,...

use GaussAdjoint in automatic sensealg

1

Triggering CI to see which tests this causes to fail.

I can't find the control flow for SDEProblem in: https://github.com/SciML/SciMLSensitivity.jl/blob/master/src/concrete_solve.jl#L76-L184 From reading the docs, `QuadratureAdjoint` should not be chosen for SDE?

CI Trigger

4

This hangs indefinitely for me locally on windows. Let us see if CI has the same issue. ```jl julia> sensealg = SciMLSensitivity.setvjp(sensealg, SciMLSensitivity.inplace_vjp(fwd_sol.prob, fwd_sol.prob.u0, fwd_sol.prob.p, true)) ┌ Warning: Using fallback...