KSLCore/ksl.app.config.optimization/PenaltyFunctionSpec

PenaltyFunctionSpec

@Serializable

sealed class PenaltyFunctionSpec(source)

Serializable selection of a penalty function used by an optimization problem to penalize constraint violations.

Mirrors the two engine implementations of ksl.simopt.problem.PenaltyFunctionIfc that are constructed from plain data and therefore can round-trip through JSON/TOML:

WithMemory — mirrors ksl.simopt.problem.PenaltyFunctionWithMemory, whose memory factor dampens stochastic noise by scaling with the inverse square root of the response sample count;
DynamicPolynomial — mirrors ksl.simopt.problem.DynamicPolynomialPenalty, a polynomial penalty that grows with both the violation magnitude and the iteration counter.

The ksl.simopt.problem.PenaltyFunctionIfc interface itself is a fun interface and so cannot be persisted in general; only the two concrete data-only implementations have a serializable representation.

Sealed-class polymorphic serialization is used: the JSON/TOML output carries a "type" discriminator with values "withMemory" or "dynamicPolynomial".

Inheritors

WithMemory

DynamicPolynomial

Constructors

PenaltyFunctionSpec

protected constructor()

Types

DynamicPolynomial

@Serializable

@SerialName(value = "dynamicPolynomial")

data class DynamicPolynomial(val basePenalty: Double = 100.0, val iterationExponent: Double = 1.0, val violationExponent: Double = 2.0) : PenaltyFunctionSpec

Mirrors ksl.simopt.problem.DynamicPolynomialPenalty.

WithMemory

@Serializable

@SerialName(value = "withMemory")

data class WithMemory(val basePenalty: Double = 100.0, val iterationExponent: Double = 1.0, val violationExponent: Double = 2.0) : PenaltyFunctionSpec

Mirrors ksl.simopt.problem.PenaltyFunctionWithMemory.