kumulant
com.eignex.kumulant/stat/regression/glm/HierarchicalBayesianRegression

HierarchicalBayesianRegression

class HierarchicalBayesianRegression(val featureSize: Int, val link: Link = Link.Identity, val biasPriorVariance: Double = 1.0, val concurrency: Concurrency = Concurrency.None, initialPriorVariance: Double = 1.0, initialPriorMean: DenseVector? = null, initialPriorCovariance: DenseMatrix? = null)(source)

Manager for a population of BayesianRegressionStat instances that share an empirical-Bayes prior. New instances inherit the current populationPrior; periodic refit re-fits the prior from the current per-instance posteriors. Cross-instance transfer happens through that prior; older instances keep accumulating their own state, but freshly created instances borrow from the population's collective experience.

Not thread-safe: callers should serialise createInstance / untrack / refit externally if used concurrently. Individual instances inherit the concurrency level chosen here for their own update path.

Typical lifecycle:

val pop = HierarchicalBayesianRegression(featureSize = 8)
val instanceA = pop.createInstance()  // seeded from initial isotropic prior
// ... feed observations to instanceA ...
val instanceB = pop.createInstance()  // still using initial prior
pop.refit()                           // population prior now reflects instanceA's evidence
val instanceC = pop.createInstance()  // benefits from instanceA's data via prior
Content copied to clipboard

Constructors

HierarchicalBayesianRegression

constructor(featureSize: Int, link: Link = Link.Identity, biasPriorVariance: Double = 1.0, concurrency: Concurrency = Concurrency.None, initialPriorVariance: Double = 1.0, initialPriorMean: DenseVector? = null, initialPriorCovariance: DenseMatrix? = null)(source)

Properties

biasPriorVariance

val biasPriorVariance: Double(source)

Bias prior variance forwarded to each instance's constructor; not refitted.

concurrency

val concurrency: Concurrency(source)

Concurrency level forwarded to each instance.

featureSize

val featureSize: Int(source)

Feature dimensionality of every managed instance.

instanceCount

val instanceCount: Int(source)

Number of instances currently contributing to refit.

populationPrior

var populationPrior: PopulationPrior(source)

Population prior used to seed new instances. Initially isotropic (N(0, initialPriorVariance * I)) or seeded from caller-supplied mean/covariance; call refit to update it from the current per-instance posteriors.

Functions

createInstance

fun createInstance(): BayesianRegressionStat(source)

Allocate a fresh BayesianRegressionStat seeded with the current populationPrior.

refit

fun refit()(source)

Refit populationPrior from the current per-instance posteriors. No-op if no instances are tracked. Existing instances keep their state; only future createInstance calls see the updated prior.

untrack

fun untrack(instance: BayesianRegressionStat)(source)

Remove instance from population tracking. Future refit calls will not see its posterior. Useful when an instance is retired or its data has gone stale.