# Trouble using ForwardDiff on a function with a cache vector and a custom struct

**URL:** <https://discourse.julialang.org/t/trouble-using-forwarddiff-on-a-function-with-a-cache-vector-and-a-custom-struct/63420>\
**Category:** General Usage\
**Created:** [June 23, 2021, 7:31am UTC](https://discourse.julialang.org/t/trouble-using-forwarddiff-on-a-function-with-a-cache-vector-and-a-custom-struct/63420 "2021-06-23T07:31:50Z")\
**Posts on this page:** 1\
**Showing post:** 2

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [July 2, 2021, 4:04pm UTC](https://discourse.julialang.org/t/trouble-using-forwarddiff-on-a-function-with-a-cache-vector-and-a-custom-struct/63420/2 "2021-07-02T16:04:39Z")

</div>

Two things here. First of all, I think it’s much more sane to just use a callable struct here, rather than embedding the anonymous function into the struct. That will give you a single source of truth on the parameters. But that doesn’t fix the issue. The issue was ` x -> get_tmp(t2.gparam,x)`, the `x` you’re passing in was `1.5`, so it was never a dual number. You need to make sure `get_tmp` is triggered by the duals, so you need to pass it `p` so that way it knows when it’s differentiating.

So a fixed version is:

```julia
using DiffEqBase: get_tmp, dualcache, DiffCache
using ForwardDiff

mutable struct Modelparameters{T,chunk_size}
    gparam::DiffCache{Array{T,1},Array{ForwardDiff.Dual{nothing,T,chunk_size},1}}
end

function (t2::Modelparameters)(x,p=nothing)
    if p !== nothing
        tmp = get_tmp(t2.gparam,p)
        tmp .= p
    else
        tmp = get_tmp(t2.gparam,x)
    end
    tmp[1]*(x-1)^tmp[2]
end

using ForwardDiff
using DiffEqBase: get_tmp, dualcache

function init()
    t1 = dualcache([1.5e-6, 1.5])
    t2 = Modelparameters(t1)
    return t2
end

t2 = init()
result1 = t2(2)
newp = [2e-6, 1.5]
changeparams!(t2, newp)
result2 = t2(2)

f = p -> t2(1.5,p)
f2 = p -> p[1]*(1.5-1)^p[2]
result3 = f([5e-6, 1.5])
result4 = f2([5e-6, 1.5])

gradient1 = ForwardDiff.gradient(f, [5e-6, 1.5])
gradient2 = ForwardDiff.gradient(f2, [5e-6, 1.5])

```

---

_[View the full topic](https://discourse.julialang.org/t/trouble-using-forwarddiff-on-a-function-with-a-cache-vector-and-a-custom-struct/63420)._