HMath.Optimization.Global¶
Module index · Numerical conventions
Description
Seeded bounded CPU global search. Bounds are finite inclusive boxes. Global optimality cannot be certified by these stochastic algorithms: the result's convergence flag denotes population/step collapse only. Exhausted generations return the best point with optConverged=False. Random state is explicit and deterministic; no system entropy or global generator is used.
On this page¶
| API | Purpose |
|---|---|
| GlobalMethod | Seeded bounded stochastic search strategy. |
| SimulatedAnnealing | Declaration and type contract below. |
| DifferentialEvolution | Declaration and type contract below. |
| GeneticAlgorithm | Declaration and type contract below. |
| ParticleSwarm | Declaration and type contract below. |
| CMAES | Declaration and type contract below. |
| globalMinimize | Run a reproducible seeded stochastic algorithm in a finite box. size>=4; budget exhaustion returns the best finite incumbent with convergence=False. |
| basinHopping | Basin hopping uses genuine local Nelder–Mead descents followed by seeded Metropolis transitions. |
Documentation
GlobalMethod
data GlobalMethod Source #
Seeded bounded stochastic search strategy. A successful finite run does not certify global optimality.
Instances
| Show GlobalMethod Source # | |
Defined in HMath.Optimization.Global Methods showsPrec :: Int -> GlobalMethod -> ShowS # show :: GlobalMethod -> String # showList :: [GlobalMethod] -> ShowS # | |
| Eq GlobalMethod Source # | |
Defined in HMath.Optimization.Global | |
globalMinimize
globalMinimize :: GlobalMethod -> Int -> Word64 -> OptimizeOptions -> [(Double, Double)] -> Objective -> Either MathError OptimizeResult Source #
Run a reproducible seeded stochastic algorithm in a finite box. size>=4; budget exhaustion returns the best finite incumbent with convergence=False.
basinHopping
basinHopping :: Word64 -> Double -> Int -> OptimizeOptions -> [(Double, Double)] -> Objective -> [Double] -> Either MathError OptimizeResult Source #
Basin hopping uses genuine local Nelder–Mead descents followed by seeded Metropolis transitions. Bounds are enforced by rejecting out-of-box trials.