Use of default_options in fminbox.jl?

[Yuan-Ru-Lin]

For unconstrained optimizer, there is a function default_options that can be extended to set default options of one's interests.

Optim.jl/src/multivariate/optimize/interface.jl

Lines 19 to 27 in 1f1258c

	default_options(method::AbstractOptimizer) = NamedTuple()
	function add_default_opts!(opts::Dict{Symbol, Any}, method::AbstractOptimizer)
	for newopt in default_options(method)
	if !haskey(opts, newopt[1])
	opts[newopt[1]] = newopt[2]
	end
	end
	end

An example is as follows.

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

The reason this is possible is because there are relevant handling in the most generic function:

Optim.jl/src/multivariate/optimize/interface.jl

Lines 83 to 92 in 1f1258c

	function optimize(f, initial_x::AbstractArray; inplace = true, autodiff = :finite, kwargs...)
	method = fallback_method(f)
	checked_kwargs, method = check_kwargs(kwargs, method)
	d = promote_objtype(method, initial_x, autodiff, inplace, f)
	add_default_opts!(checked_kwargs, method)
	options = Options(; checked_kwargs...)
	optimize(d, initial_x, method, options)
	end

However, this is not true in the case of optimize that takes Fminbox as an argument:

Optim.jl/src/multivariate/solvers/constrained/fminbox.jl

Lines 259 to 270 in 1f1258c

	function optimize(f,
	l::AbstractArray,
	u::AbstractArray,
	initial_x::AbstractArray,
	F::Fminbox = Fminbox(),
	options::Options = Options(); inplace = true, autodiff = :finite)
	if f isa NonDifferentiable
	f = f.f
	end
	od = OnceDifferentiable(f, initial_x, zero(eltype(initial_x)); autodiff = autodiff)
	optimize(od, l, u, initial_x, F, options)
	end