API

This section documents all functions not documented elsewhere

SemioticOpt.iteration — Function

iteration(f::Function, a::GradientDescent)

One iteration of a on the function f.

This function is unexported.

source

iteration(f::Function, a::ProjectedGradientDescent)

Apply a to f, projected based on a.t.

source

iteration(obj::Function, alg::PairwiseGreedyOpt)

One iteration of the pairwise greedy optimization algorithm alg for objective obj.

Example

julia> using SemioticOpt
julia> using LinearAlgebra
julia> f(x, ixs, a, b) = -((a[ixs] .* x) ./ (x .+ b[ixs])) |> sum
julia> aa = Float64[1, 1, 1000, 1]
julia> bb = Float64[1, 1, 1, 1]
julia> f(x, ixs) = f(x, ixs, aa, bb)
julia> function makepgd(v)
           return ProjectedGradientDescent(;
               x=v,
               η=1e-1,
               hooks=[StopWhen((a; kws...) -> norm(SemioticOpt.x(a) - kws[:z]) < 1.0)],
               t=v -> σsimplex(v, 1)  # Project onto unit-simplex
           )
       end
julia> alg = PairwiseGreedyOpt(;
           kmax=4,
           x=zeros(4),
           xinit=zeros(4),
           f=f,
           a=makepgd,
           hooks=[StopWhen((a; kws...) -> kws[:f](kws[:z]) ≥ kws[:f](SemioticOpt.x(a)))]
       )
julia> c = 0.1  # per non-zero cost
julia> selection = x -> f(x, 1:length(x)) + c * length(SemioticOpt.nonzeroixs(x))
julia> z = SemioticOpt.iteration(selection, alg)
4-element Vector{Float64}:
 0.0
 0.0
 1.0
 0.0

source

SemioticOpt.f — Function

Getter for the stop-function.

source

Get the function to record data

source

f(g::PairwiseGreedyOpt)

The objective function of the inner loop.

source

SemioticOpt.η — Function

η(g::GradientDescent)

The learning rate/step size.

source

SemioticOpt.hooks — Function

hooks(g::GradientDescent)

The hooks used by the algorithm.

source

hooks(g::PairwiseGreedyOpt)

The hooks used by the algorithm.

source

SemioticOpt.λ — Function

Get the λ of h.

source

SemioticOpt.maybeminimize! — Function

maybeminimize!(f::Function, a::OptAlgorithm, op::Function)

Minimize f using a, which calls op for updating a.x.

This function may be in-place. Don't use it directly unless you know what you're doing. This function is unexported.

Warning

If you don't provide any hook with the IsStoppingCondition trait, this will loop forever.

source

SemioticOpt.postiteration — Function

postiterationhook(
    hs::H, a::OptAlgorithm, z::AbstractVector{T}; locals...
) where {H<:Hooks,T<:Real}

Run hooks that the code should execute after SemioticOpt.iteration.

source

SemioticOpt.shouldstop — Function

shouldstop(hs::Hooks, a::OptAlgorithm; locals...)

Map over the hooks hs to check for stopping conditions.

This performs an OR operation if there are multiple hooks that have the SemioticOpt.IsStoppingCondition trait.

source

SemioticOpt.x — Function

x(g::GradientDescent)
x(g::GradientDescent, v)

The current best guess for the solution. If using the setter, v is the new value.

The setter is not in-place. See SemioticOpt.x!.

source

x(g::ProjectedGradientDescent)
x(g::ProjectedGradientDescent, v)

The current best guess for the solution. If using the setter, v is the new value.

The setter is not in-place. See SemioticOpt.x!.

source

x(g::PairwiseGreedyOpt)
x(g::PairwiseGreedyOpt, v)

The current best guess for the solution. If using the setter, v is the new value.

The setter is not in-place. See SemioticOpt.x!.

source

SemioticOpt.x! — Function

x!(g::GradientDescent, v)

In-place setting of g.x to v

See SemioticOpt.x.

source

x!(g::ProjectedGradientDescent, v)

In-place setting of a.g.x to v

See SemioticOpt.x.

source

x!(g::PairwiseGreedyOpt, v)

In-place setting of g.x to v

See SemioticOpt.x.

source

SemioticOpt.x₀ — Function

Get the anchor of h.

source

SemioticOpt.name — Function

Get the name of the h.

source

SemioticOpt.data — Function

Get the data of the h.

source