kumulant/com.eignex.kumulant.bandit.univariate/MeanArm

MeanArm

@Serializable

@SerialName(value = "MeanArm")

data class MeanArm(val priorMean: Double = 1.0, val priorWeight: Double = 0.01) : Arm<WeightedMeanResult> (source)

Single-moment arm; tracks the running mean but not variance. The right pick when the likelihood's sufficient statistic is one running sum (or equivalently a running mean × count):

Poisson reward (count data): pair with PoissonGammaPosterior.
Geometric reward (trials until success): pair with GeometricBetaPosterior.
Exponential reward (inter-arrival time): pair with ExponentialGammaPosterior.
Gamma reward with known shape: pair with GammaScalePosterior.

The prior is a pseudo-observation of value priorMean with weight priorWeight. Tiny priorWeight (default 0.01) makes the prior a soft suggestion that gets washed out in the first few real observations.

Constructors

MeanArm

constructor(priorMean: Double = 1.0, priorWeight: Double = 0.01)

Types

Companion

object Companion

Factory entry-point for MeanArm (host for warmStart).

Properties

priorMean

val priorMean: Double

Prior mean reward, seeded as a pseudo-observation of weight priorWeight.

priorWeight

val priorWeight: Double

Pseudo-weight of the prior seed; smaller = weaker prior.

Functions

createStat

open override fun createStat(): SeriesStat<WeightedMeanResult>

Allocate a fresh per-arm accumulator already seeded with this arm's prior pseudo-counts.

encode

open fun encode(value: Double): Double

Map a raw observation onto the scale the stat accumulates. Identity by default; LogNormalArm overrides with ln so the underlying stat tracks the log-reward and the Normal-Gamma posterior fits the log-normal generative model.