Benoît Pasquier

Thank you for trying to enlighten me. However, I am still quite lost and overwhelmed with the links you provided. Could you comment on my understanding and my suggestion? Right...

I'm sorry to insist, but I still think this could improve [ForwardDiff.jl](github.com/JuliaDiff/ForwardDiff.jl)'s efficiency when dealing with `\` and factorizations. I have implemented my own factorization for my use case, but...

Yes I try to follow what's happening with ForwardDiff2.jl :) FYI I also made [DualMatrixTools.jl](https://github.com/briochemc/DualMatrixTools.jl) and [HyperDualMatrixTools.jl](https://github.com/briochemc/HyperDualMatrixTools.jl) too a year or so back, if that's any interest to you.

FWIW I realized from a slack comment that one should do ```julia directionalderivative(f, x, y) = ForwardDiff.derivative(λ -> f(x .+ λ .* y), 0.0) ``` instead. Maybe that's as good...

[ForwardDiff.jl](github.com/JuliaDiff/ForwardDiff.jl) applies to functions. So in your case, you would need to define a function `f` such that `magnitude` is equal to `f` applied to all the values in `t`,...

You do not need any packages for that, you just need to apply finite differences with your time steps: ```julia velocity = diff(magnitude) ./ diff(t) ```

I guess this issue can be closed now?

I just tried to play with taking the derivative of a simple Unitful-embedded function: ```julia julia> using Unitful, ForwardDiff julia> foo(x) = x / 1.0u"s" foo (generic function with 1...

@astrozot does that work for you? ```julia ForwardDiff.derivative(λ -> f(x .+ λ), 0.0) ``` Unless any call to `ForwardDiff.derivative` is bad? [Edit: I guess @KristofferC's reply means it's bad 😥]

+1 Also, it would be nice to have a little example for people like me who already had saved data in JLD format (which versions 0.7 and 1.0 cannot access...