SimultaneousPerturbationStochasticApproximation

using SimultaneousPerturbationStochasticApproximation
f(x) = sum(abs2, x)
result = minimize(f, lower = fill(-10, 10), upper = fill(10, 10), maxfevals = 10^5)

1-element Array{Any,1}:
 (x = [5.709728947263545e-21, 2.5332645667262345e-21, 2.037385585780084e-20, 1.1949949632333856e-20, -2.41682064852466e-20, 1.7430842882710955e-20, -1.9709712546525954e-20, -9.674114437961352e-21, 1.3385675841053994e-20, -1.2440646770159904e-20], f = 1.8730950287147333e-6, evals = 8201)

SimultaneousPerturbationStochasticApproximation.ConvergenceWelchTest — Method

ConvergenceWelchTest(; N = 1000, K = 25, T = Float64, N_positivetests = 3)

source

SimultaneousPerturbationStochasticApproximation.LearningRateUpdate — Method

LearningRateUpdate(lower, upper; pi_a = .7, N = 1000, factor = 1.5)

source

SimultaneousPerturbationStochasticApproximation.NoInitializer — Type

NoInitializer()

source

SimultaneousPerturbationStochasticApproximation.RandomInitializer — Type

RandomInitializer()

source

SimultaneousPerturbationStochasticApproximation.SPSA — Method

SPSA(; lower, upper,
       heuristic = SimpleHeuristic(elementwise = true,
                                   update = LearningRateUpdate(lower, upper)),
       initializer = SobolInitializer(restarts = 1, N = 100),
       convergencetest = ConvergenceWelchTest(),
       pi_max = .1,
       init = lower)

source

SimultaneousPerturbationStochasticApproximation.SimpleHeuristic — Method

SimpleHeuristic(; elementwise = false, A = 500, c = 0., c_min = 1e-3,
                  a_max = 1e4, α = .602, γ = .101, ninit = 100,
                  update = nothing)

update could also be LearningRateUpdate.

source

SimultaneousPerturbationStochasticApproximation.minimize! — Method

minimize!(spsa::SPSA, f; kwargs...) = minimize!(Random.GLOBAL_RNG, spsa, f; kwargs...)

source

SimultaneousPerturbationStochasticApproximation.minimize! — Method

minimize!(rng::Random.AbstractRNG,
               spsa::SPSA{<:Any,<:Any,<:Any,T},
               f;
               callback = () -> nothing,
               maxfevals,
               gamma_fhat = .9, # just for tracking the learning progress
               verbose = true) where T

source

SimultaneousPerturbationStochasticApproximation.minimize — Method

minimize(f; callback = () -> nothing, maxfevals, verbose = true, kwargs...)

kwargs are passed to SPSA.

source