# Optim.jl - No default objective type for IPNewton

**URL:** <https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036>\
**Category:** Optimization (Mathematical)\
**Tags:** optim\
**Created:** [April 11, 2019, 2:32pm UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036 "2019-04-11T14:32:34Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![ferrolho](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/ferrolho/32/213665_2.png) [@ferrolho](https://discourse.julialang.org/u/ferrolho)\
**Post date:** [April 11, 2019, 2:32pm UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036/1 "2019-04-11T14:32:34Z")

</div>

Please consider the following code snippet using `Optim.jl`:

```julia
using Optim

dof = 7

fun(x) = 0.0; x0 = fill(0.1, dof)
df = TwiceDifferentiable(fun, x0)

lx = fill(-1.2, dof); ux = fill(+1.2, dof)
dfc = TwiceDifferentiableConstraints(lx, ux)

# res = optimize(df, dfc, x0, IPNewton())
res = optimize(df, dfc, x0, IPNewton(); autodiff=:forward)

```

If I call `optimize()` without the `autodiff=:forward` flag, the snippet above works. However, if I use set `autodiff=:forward`, I get the following error:

```julia
No default objective type for IPNewton{typeof(Optim.backtrack_constrained_grad),Symbol}(Optim.backtrack_constrained_grad, :auto, false) and (TwiceDifferentiable{Float64,Array{Float64,1},Array{Float64,2},Array{Float64,1}}(fun, getfield(NLSolversBase, Symbol("#g!#44")){typeof(fun),DiffEqDiffTools.GradientCache{Nothing,Nothing,Nothing,Val{:central},Float64,Val{true}}}(fun, DiffEqDiffTools.GradientCache{Nothing,Nothing,Nothing,Val{:central},Float64,Val{true}}(nothing, nothing, nothing)), getfield(NLSolversBase, Symbol("#fg!#45")){typeof(fun)}(fun, Core.Box(getfield(NLSolversBase, Symbol("#g!#44")){typeof(fun),DiffEqDiffTools.GradientCache{Nothing,Nothing,Nothing,Val{:central},Float64,Val{true}}}(fun, DiffEqDiffTools.GradientCache{Nothing,Nothing,Nothing,Val{:central},Float64,Val{true}}(nothing, nothing, nothing)))), getfield(NLSolversBase, Symbol("#h!#46")){typeof(fun)}(fun), 0.0, [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0], [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], [1], [1], [1]), TwiceDifferentiableConstraints{getfield(NLSolversBase, Symbol("##88#91")),getfield(NLSolversBase, Symbol("##89#92")),getfield(NLSolversBase, Symbol("##90#93")),Float64}(getfield(NLSolversBase, Symbol("##88#91"))(), getfield(NLSolversBase, Symbol("##89#92"))(), getfield(NLSolversBase, Symbol("##90#93"))(), ConstraintBounds:
  Variables:
    x[1]≥-1.2, x[1]≤1.2, x[2]≥-1.2, x[2]≤1.2, x[3]≥-1.2, x[3]≤1.2, x[4]≥-1.2, x[4]≤1.2, x[5]≥-1.2, x[5]≤1.2, x[6]≥-1.2, x[6]≤1.2, x[7]≥-1.2, x[7]≤1.2
  Linear/nonlinear constraints:)).

Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] promote_objtype(::IPNewton{typeof(Optim.backtrack_constrained_grad),Symbol}, ::Array{Float64,1}, ::Symbol, ::Bool, ::TwiceDifferentiable{Float64,Array{Float64,1},Array{Float64,2},Array{Float64,1}}, ::TwiceDifferentiableConstraints{getfield(NLSolversBase, Symbol("##88#91")),getfield(NLSolversBase, Symbol("##89#92")),getfield(NLSolversBase, Symbol("##90#93")),Float64}) at /home/henrique/.julia/packages/Optim/Agd3B/src/multivariate/optimize/interface.jl:37
 [3] #optimize#88(::Bool, ::Symbol, ::Function, ::TwiceDifferentiable{Float64,Array{Float64,1},Array{Float64,2},Array{Float64,1}}, ::TwiceDifferentiableConstraints{getfield(NLSolversBase, Symbol("##88#91")),getfield(NLSolversBase, Symbol("##89#92")),getfield(NLSolversBase, Symbol("##90#93")),Float64}, ::Array{Float64,1}, ::IPNewton{typeof(Optim.backtrack_constrained_grad),Symbol}, ::Optim.Options{Float64,Nothing}) at /home/henrique/.julia/packages/Optim/Agd3B/src/multivariate/optimize/interface.jl:121
 [4] (::getfield(Optim, Symbol("#kw##optimize")))(::NamedTuple{(:autodiff,),Tuple{Symbol}}, ::typeof(optimize), ::TwiceDifferentiable{Float64,Array{Float64,1},Array{Float64,2},Array{Float64,1}}, ::TwiceDifferentiableConstraints{getfield(NLSolversBase, Symbol("##88#91")),getfield(NLSolversBase, Symbol("##89#92")),getfield(NLSolversBase, Symbol("##90#93")),Float64}, ::Array{Float64,1}, ::IPNewton{typeof(Optim.backtrack_constrained_grad),Symbol}, ::Optim.Options{Float64,Nothing}) at ./none:0 (repeats 2 times)
 [5] top-level scope at In[51]:1

```

Can somebody help me understand why? Is it perhaps because AD is not supported for _IPNewton_ with multiple constraints? Thanks!

---

<div class="post-metadata">

**Author:** ![pkofod](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pkofod/32/2179_2.png) [@pkofod](https://discourse.julialang.org/u/pkofod)\
**Post date:** [April 11, 2019, 4:43pm UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036/2 "2019-04-11T16:43:32Z")

</div>

Its probably an oversight. What version of optim?

---

<div class="post-metadata">

**Author:** ![ferrolho](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/ferrolho/32/213665_2.png) [@ferrolho](https://discourse.julialang.org/u/ferrolho)\
**Post date:** [April 11, 2019, 6:40pm UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036/3 "2019-04-11T18:40:23Z")

</div>

Thanks for your reply! `(v1.1) pkg> st` shows `[429524aa] Optim v0.18.1`. What do you mean by oversight?

---

<div class="post-metadata">

**Author:** ![pkofod](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pkofod/32/2179_2.png) [@pkofod](https://discourse.julialang.org/u/pkofod)\
**Post date:** [April 11, 2019, 7:18pm UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036/4 "2019-04-11T19:18:14Z")

</div>

I mean that I probably forgot to think of this edge case 🙂 Can you open an issue on [https://github.com/JuliaNLSolvers/Optim.jl/issues/new](https://github.com/JuliaNLSolvers/Optim.jl/issues/new) ? I know you were already sent here from slack.

Are you trying to specify that the TwiceDifferentiable function should be differentiated using forwarddiff? If so, that should be specified in the `TwiceDifferentiable(fun, x0; autodiff=:forwarddiff)` part of the code, or you should just pass `fun` to `optimize` (though that method may also not exist).

---

<div class="post-metadata">

**Author:** ![ferrolho](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/ferrolho/32/213665_2.png) [@ferrolho](https://discourse.julialang.org/u/ferrolho)\
**Post date:** [April 12, 2019, 10:03am UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036/5 "2019-04-12T10:03:13Z")

</div>

Sure! I have opened an issue here: [No default objective type for IPNewton · Issue #711 · JuliaNLSolvers/Optim.jl · GitHub](https://github.com/JuliaNLSolvers/Optim.jl/issues/711).

Actually, I’d like to use _forwarddiff_ for both `df` and `dfc`, since in my more complex application I want to have something like `dfc = TwiceDifferentiableConstraints(con_c!, lx, ux, lc, uc; autodiff=:forward)`. Unfortunately this does not work, currently.

* * *

On a slightly different note, I am also a bit confused by the [`NDifferentiable` instances](https://github.com/JuliaNLSolvers/NLSolversBase.jl#ndifferentiable):

> The words in front of `Differentiable` in the type names ( `Non` , `Once` , `Twice` ) are not meant to indicate a specific classification of the function as such (a `OnceDifferentiable` might be constructed for an infinitely differentiable function), but signals to an algorithm if the correct functions have been constructed or if automatic differentiation should be used to further differentiate the function.

So, is there a difference between

```julia
df = TwiceDifferentiable(fun, x0; autodiff=:forwarddiff)

```

and

```julia
df = NonDifferentiable(fun, x0; autodiff=:forwarddiff)

```

or even

```julia
df = NonDifferentiable(fun, x0)

```

?

Shouldn’t the latter be the preferable syntax for a differentiable function but for which a user did not supply the Jacobian or Hessian manually? Maybe I should start a new topic for this question alone…

---

<div class="post-metadata">

**Author:** ![pkofod](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pkofod/32/2179_2.png) [@pkofod](https://discourse.julialang.org/u/pkofod)\
**Post date:** [April 12, 2019, 10:19am UTC](https://discourse.julialang.org/t/optim-jl-no-default-objective-type-for-ipnewton/23036/6 "2019-04-12T10:19:46Z")

</div>

The quick answer is: no. \*NDifferentiable is an implementation detail, and few users should want to touch it at all. I generally want people to use

```julia
optimize(f, g!, x, method, options)

```

The only issue is with people who are obsessed with cache arrays because they’re doing similar optimizations over and over again. The only problem is that the constrained optimization interface isn’t as well-developed, so there are probably quite a few cases where the interface that should work simply doesn’t.