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

RegressionAuditLeaf

class RegressionAuditLeaf<Row>(val arm: SeriesStat<WeightedVarianceResult>, val candidates: List<Split<Row>>, val pos: List<SeriesStat<WeightedVarianceResult>>, val neg: List<SeriesStat<WeightedVarianceResult>>) : RegressionLeafNode<Row> (source)

Leaf that tracks per-candidate pos/neg stats. When a candidate clears the Hoeffding- bound test, this leaf is replaced by a RegressionSplitNode. The candidate subset is per-leaf ; picked at leaf birth; so mtry-style random subspace selection lives at the leaf level.

Constructors

RegressionAuditLeaf

constructor(arm: SeriesStat<WeightedVarianceResult>, candidates: List<Split<Row>>, pos: List<SeriesStat<WeightedVarianceResult>>, neg: List<SeriesStat<WeightedVarianceResult>>)(source)

Properties

arm

open override val arm: SeriesStat<WeightedVarianceResult>(source)

The leaf's weighted-variance accumulator.

candidates

val candidates: List<Split<Row>>(source)

Candidate splits being evaluated at this leaf.

neg

val neg: List<SeriesStat<WeightedVarianceResult>>(source)

Per-candidate accumulator for observations that route false.

observationsSinceLastCheck

val observationsSinceLastCheck: AtomicLong(source)

Counter throttling audit work to every Nth observation. Atomic so concurrent updaters don't lose ticks; the split decision is taken by the thread that crosses the period boundary, under the tree's split lock.

pos

val pos: List<SeriesStat<WeightedVarianceResult>>(source)

Per-candidate accumulator for observations that route true.

Functions

aggregate

fun RegressionNode<*>.aggregate(): WeightedVarianceResult

Public per-node subtree aggregate over all observations routed through this node (leaves: the arm snapshot; splits: the exact Chan-merge of both children plus any RegressionSplitNode.carryover). Lets callers score internal nodes — e.g. a Thompson walk-down that picks a branch by its subtree's posterior — without the tree having to keep a live arm on every split node. Equivalent to a directly-accumulated internal arm: Chan's parallel merge is exact for weighted-variance aggregates.

findLeaf

override fun findLeaf(row: Row): RegressionLeafNode<Row>

Walk to the leaf this row resolves to.

snapshot

fun RegressionNode<VectorView>.snapshot(): TreeNodeResult

Freeze a live VectorView tree node into an immutable, serializable snapshot. Internal split aggregates are derived from the snapshotted children so the wire format stays stable even though live splits hold no arm.