In the interest of user-friendliness, would anyone know a way to overload ‘optimize’ to work conveniently for scalar input without crashing?
[please note I am not asking for opinions as to how good/bad an idea it would be - let’s assume I am willing to undertake any ‘risks’].
Of course, I can get it to work if I redefine my objective function to have one-dimensional Array argument and likewise my starting point, and then tell optimize I do not want Nelder-Mead, but let’s assume I would like to use optimize frequently and conveniently for scalar input and don’t want to go through the above steps every time. I would like (for my own use) to make the routine more user-friendly in this regard. Can anyone suggest a way for me to conveniently do this?
[PS I don’t mean the ‘Brent’s’ solution which apparently works - I want something fast]
1 Like
I am not sure you are aware of the possible pitfalls. Curiously, multivariate methods can break down in surprising ways in 1D, and can easily yield suboptimal performance. Optim also has GoldenSection(), see
https://julianlsolvers.github.io/Optim.jl/stable/#user/minimization/#minimizing-a-univariate-function-on-a-bounded-interval
That said, you can always write a wrapper like
using Optim
function univariate_optimize(f, x0, args...; kwargs...)
opt = Optim.optimize(x -> f(x[1]), [x0], args...; kwargs...)
@assert Optim.converged(opt)
Optim.minimizer(opt)[1]
end
univariate_optimize(x -> abs2(x), 1.0, BFGS(); autodiff = :forward)
You just have to make a decision about what to extract from Optim.MultivariateOptimizationResults, or write a conversion routine to Optim.UnivariateOptimizationResults.
2 Likes
Thanks - I searched but didn’t see that one.