Set default options and settings for Optim.optimize()

[Yuan-Ru-Lin]

Right now one has to type the following to perform fitting.

optimize(x -> -loglikelihood(model(x), data), x0, BFGS(), Optim.Options(extended_trace=true, callback=miunitstop))
While defining the function and giving x0 may be necessary and would be better stated explicitly, the method BFGS() and the options are not, as they should be the same unless there is a really good reason).

It is possible to wrap the call but I still want to keep it as explicit as possible. A variable for configuration would make things easier such as the PLOTS_DEFAULTS in the case of Plots.jl.

[Yuan-Ru-Lin]

Something like

mydefault(; additionals...)::Optim.Options = Optim.Options(extended_trace=true, callback=miunitstop; additionals...)

should work. This way, the call would become

optimize(x -> -loglikelihood(model(x), data), x0, BFGS(), mydefault())

while keeping flexibility to pass keyword arguments to control the options.

[Yuan-Ru-Lin]

import Optim: default_options
using Optim: BFGS
default_options(::BFGS) = (; extended_trace=true, callback=miunitstop)

This will set the default options for BFGS().

However, there is no such mechanism for Fminbox(BFGS()).

See also

https://github.com/JuliaNLSolvers/Optim.jl/blob/1f1258c8499b57c6a551ef94057604bd58f1aa31/src/multivariate/solvers/constrained/fminbox.jl#L272-L278

[Yuan-Ru-Lin]

I overrode optimize:

function MiunitOptions(; additionals...)::Optim.Options
    Optim.Options(extended_trace=true, callback=miunitstop; additionals...)
end

import Optim: optimize
function optimize(f,
                  l::AbstractArray,
                  u::AbstractArray,
                  initial_x::AbstractArray,
                  F::T) where {T <: Fminbox}
    optimize(f, l, u, initial_x, F, MiunitOptions())
end

I've opened an issue at Optim (JuliaNLSolvers/Optim.jl#1028). If this gets resolved, this can be replaced with something like default_options(::BFGS) = ...