KSLCore/ksl.app.config.optimization/SolverSpec

SolverSpec

@Serializable

Serializable selection of a simulation-optimization solver and its algorithm-specific parameters.

Each sealed variant captures the constructor parameters that are plain data; non-serializable arguments to the live solver constructors (evaluator, stream provider, problem definition, equality checkers, functional sampler/elite-size hooks) are supplied by OptimizationSolverFactory and are not part of the persisted document.

Sealed-class polymorphic serialization is used: the JSON/TOML output carries a "type" discriminator with values "stochasticHillClimbing", "simulatedAnnealing", "crossEntropy", or "rSpline".

Random-restart is not a separate sealed variant. Any algorithm can be wrapped by setting randomRestart on its variant; setting it to null (the default) disables the wrapper. This keeps the sealed hierarchy focused on algorithm choice and lets every algorithm be wrapped uniformly.

Inheritors

StochasticHillClimbing

SimulatedAnnealing

CrossEntropy

RSpline

Constructors

SolverSpec

protected constructor()

Types

CrossEntropy

@Serializable

@SerialName(value = "crossEntropy")

data class CrossEntropy(val startingPoint: Map<String, Double>? = null, val maxIterations: Int, val randomRestart: RandomRestartSpec? = null, val streamNum: Int = 0, val name: String? = null, val replicationsPerEvaluation: Int, val sampler: CESamplerSpec = CESamplerSpec.Normal(), val elitePct: Double? = null, val ceSampleSize: Int? = null) : SolverSpec

Cross-Entropy. Mirrors ksl.simopt.solvers.algorithms.CrossEntropySolver.

RSpline

@Serializable

@SerialName(value = "rSpline")

data class RSpline(val startingPoint: Map<String, Double>? = null, val maxIterations: Int, val randomRestart: RandomRestartSpec? = null, val streamNum: Int = 0, val name: String? = null, val initialNumReps: Int, val sampleSizeGrowthRate: Double, val maxNumReplications: Int) : SolverSpec

R-SPLINE. Mirrors ksl.simopt.solvers.algorithms.RSplineSolver.

SimulatedAnnealing

@Serializable

@SerialName(value = "simulatedAnnealing")

data class SimulatedAnnealing(val startingPoint: Map<String, Double>? = null, val maxIterations: Int, val randomRestart: RandomRestartSpec? = null, val streamNum: Int = 0, val name: String? = null, val replicationsPerEvaluation: Int, val temperature: TemperatureSpec = TemperatureSpec.AutoCalibrate(), val coolingSchedule: CoolingScheduleSpec, val stoppingTemperature: Double) : SolverSpec

Simulated Annealing. Mirrors ksl.simopt.solvers.algorithms.SimulatedAnnealing.

StochasticHillClimbing

@Serializable

@SerialName(value = "stochasticHillClimbing")

data class StochasticHillClimbing(val startingPoint: Map<String, Double>? = null, val maxIterations: Int, val randomRestart: RandomRestartSpec? = null, val streamNum: Int = 0, val name: String? = null, val replicationsPerEvaluation: Int) : SolverSpec

Stochastic Hill Climbing. Mirrors ksl.simopt.solvers.algorithms.StochasticHillClimber.

Properties

maxIterations

abstract val maxIterations: Int

Maximum number of main-loop iterations the solver is permitted to run; must be positive when validated.

name

abstract val name: String?

Optional human-readable solver instance name.

randomRestart

abstract val randomRestart: RandomRestartSpec?

Optional random-restart wrapper. When non-null the solver factory wraps the chosen algorithm in a ksl.simopt.solvers.algorithms.RandomRestartSolver.

startingPoint

abstract val startingPoint: Map<String, Double>?

Optional starting point for the search, keyed by decision-variable name. When null, the solver chooses its own starting point.

streamNum

abstract val streamNum: Int

Random-number stream number used to seed the solver's stochastic decisions. 0 means "next available stream".