kumulant/com.eignex.kumulant.schema/HalfSpaceTrees

HalfSpaceTrees

@Serializable

@SerialName(value = "HalfSpaceTrees")

data class HalfSpaceTrees(val featureSize: Int, val featureRanges: List<FeatureRange>, val numTrees: Int = 25, val height: Int = 8, val windowSize: Int = 250, val randomSeed: Int = 0) : VectorStatSpec<HalfSpaceTreesResult> (source)

Spec for HalfSpaceTreesStat: online ensemble of random half-space trees for multivariate anomaly scoring.

Constructors

HalfSpaceTrees

constructor(featureSize: Int, featureRanges: List<FeatureRange>, numTrees: Int = 25, height: Int = 8, windowSize: Int = 250, randomSeed: Int = 0)

Properties

featureRanges

val featureRanges: List<FeatureRange>

Per-feature value ranges used to draw random split thresholds.

featureSize

val featureSize: Int

Number of input features.

height

val height: Int

Depth of each tree; each tree has 2^height leaves.

numTrees

val numTrees: Int

Number of trees in the ensemble.

randomSeed

val randomSeed: Int

PRNG seed shared across trees.

windowSize

val windowSize: Int

Observations per window before the reference profile rotates.

Functions

filter

fun <R : Result> VectorStatSpec<R>.filter(pred: BoolExpr): VectorStatSpec<R>

Wrap this vector spec so updates are forwarded only when pred evaluates true on the full vector.

materialize

fun StatSpec.materialize(concurrency: Concurrency = Concurrency.None): Stat<*>

Construct a live stat from any StatSpec, dispatching on its modality. Useful for code paths (like StatSchemaDef.materialize) that iterate over an erased Map<String, StatSpec> and don't statically know the modality.

fun <R : Result> VectorStatSpec<R>.materialize(concurrency: Concurrency = Concurrency.None): VectorStat<R>

Construct a live VectorStat from a VectorStatSpec. See SeriesStatSpec.materialize.

minMaxScaleFeatures

fun <R : Result> VectorStatSpec<R>.minMaxScaleFeatures(dimensions: Int, targetLow: Double = 0.0, targetHigh: Double = 1.0): VectorStatSpec<R>

Element-wise min-max scale a vector spec against a hidden per-coordinate Range primary.

sample

fun <R : Result> VectorStatSpec<R>.sample(rate: Double, seed: Long): VectorStatSpec<R>

Wrap this vector spec to keep each update with probability rate; seed feeds the PRNG.

standardScaleFeatures

fun <R : Result> VectorStatSpec<R>.standardScaleFeatures(dimensions: Int): VectorStatSpec<R>

Element-wise standardise a vector spec against a hidden per-coordinate Variance primary.

throttle

fun <R : Result> VectorStatSpec<R>.throttle(every: Int): VectorStatSpec<R>

Wrap this vector spec so it only sees one in every every updates.

transformElement

fun <R : Result> VectorStatSpec<R>.transformElement(expr: ScalarExpr): VectorStatSpec<R>

Wrap this vector spec to apply expr to every element of each incoming vector before update.

transformVector

fun <R : Result> VectorStatSpec<R>.transformVector(expr: VectorExpr): VectorStatSpec<R>

Wrap this vector spec so each incoming vector is remapped through expr before update.

weightBy

fun <R : Result> VectorStatSpec<R>.weightBy(expr: ScalarExpr): VectorStatSpec<R>

Wrap this vector spec so every update's weight is multiplied by expr.eval(0, 0, vec).

windowed

fun <R : Result> VectorStatSpec<R>.windowed(durationMillis: Long, slices: Int = 10): VectorStatSpec<R>

Wrap this vector spec in a sliding time window of durationMillis split into slices buckets.

withWeight

fun <R : Result> VectorStatSpec<R>.withWeight(weight: Double): VectorStatSpec<R>

Wrap this vector spec so every update applies the per-observation weight multiplier.

HalfSpaceTrees

constructor(featureSize: Int, featureRanges: List<FeatureRange>, numTrees: Int = 25, height: Int = 8, windowSize: Int = 250, randomSeed: Int = 0)(source)