com.eignex.kumulant/stat/regression/tree/ThompsonTreePosterior

ThompsonTreePosterior

data class ThompsonTreePosterior(val priorWeight: Double = 1.0, val priorVariance: Double = 1.0) : TreePosterior(source)

Thompson sampling over the leaf's Normal-Gamma posterior. Given the leaf's pseudo- count n, sample mean m, and sample variance v, draws are mu ~ N(m, exploration * v / max(n, 1)); the posterior on the leaf mean assuming a Normal-Gamma conjugate. The variance is blended with priorVariance on a priorWeight pseudo-count, so a leaf holding one observation - whose sample variance is exactly zero - still has spread. exploration = 0.0 collapses to the leaf mean.

Constructors

ThompsonTreePosterior

constructor(priorWeight: Double = 1.0, priorVariance: Double = 1.0)(source)

Properties

priorVariance

val priorVariance: Double(source)

Prior variance applied when the leaf has effectively no signal.

priorWeight

val priorWeight: Double(source)

Pseudo-count added to the leaf's totalWeights to avoid divide-by-zero on empty leaves.

Functions

evaluate

open override fun evaluate(snapshot: TreeRegressionResult, x: F64VectorView, rng: Random, exploration: Double = 1.0): Double(source)

Score a query point x under the regression snapshot. exploration controls the posterior-variance scale (Thompson) or the UCB width (LinUcb-style); 0.0 collapses to the point estimate.