Package dev.frostlake.functions.aggregate

Class HllAccumulator

java.lang.Object
dev.frostlake.functions.aggregate.HllAccumulator
All Implemented Interfaces:
AggregateFunction.Accumulator, MultiArgumentAccumulator
public class HllAccumulator extends Object implements AggregateFunction.Accumulator, MultiArgumentAccumulator
HyperLogLog sketch for APPROX_COUNT_DISTINCT (Flajolet et al.), with linear counting in the small-cardinality range. Replaces the previous approach — an exact HashSet of 32-bit String.hashCode() values — which could UNDERCOUNT whenever two distinct values collided on hashCode (e.g. "Aa" and "BB"). Each value's string form is hashed to 64 bits, so distinct values are (with overwhelming probability) counted separately. With p = 14 registers the standard error is ~0.8%, and linear counting keeps cardinalities up to ~40k (near-)exact.

  • Constructor Summary

    Constructors
    Constructor
    Description
    HllAccumulator()
     

  • Method Summary

    Modifier and Type
    Method
    Description
    void
    accumulate(Object value)
    Folds one row's argument value into the running state.
    void
    accumulate(List<Object> argumentValues)
    A multi-argument call counts distinct TUPLES, and a tuple with any NULL member is skipped whole — live counts (1.5, 7) once over {(1.5, 7), (1.5, NULL), (NULL, 7), (NULL, NULL), (1.5, 7)}.
    Object
    getResult()
    The aggregate's result over everything accumulated so far.
    void
    merge(AggregateFunction.Accumulator other)
    Folds another accumulator's state into this one, combining two partial aggregations of the same group.
    void
    reset()
    Clears the state back to the empty-group starting point so the accumulator can be reused.

    Methods inherited from class java.lang.Object

    clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait