org.apache.flink.table.runtime.functions.aggregate.hyperloglog

Class HyperLogLogPlusPlus

java.lang.Object

org.apache.flink.table.runtime.functions.aggregate.hyperloglog.HyperLogLogPlusPlus

public class HyperLogLogPlusPlus
extends Object

The implement of HyperLogLogPlusPlus is inspired from Apache Spark.

Constructor Summary

Constructors

Constructor and Description
HyperLogLogPlusPlus(double relativeSD)

Method Summary

Modifier and Type	Method and Description
double	estimateBias(double e) Estimate the bias using the raw estimates with their respective biases from the HLL++ appendix.
int	getNumWords()
void	merge(HllBuffer buffer1, HllBuffer buffer2) Merge the HLL buffers by iterating through the registers in both buffers and select the maximum number of leading zeros for each register.
long	query(HllBuffer buffer) Compute the HyperLogLog estimate.
double	trueRsd() The rsd of HLL++ is always equal to or better than the rsd requested.
void	updateByHashcode(HllBuffer buffer, long hash) Update the HLL++ buffer.

Methods inherited from class java.lang.Object

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

Constructor Detail

HyperLogLogPlusPlus

public HyperLogLogPlusPlus(double relativeSD)

Method Detail

getNumWords

public int getNumWords()

updateByHashcode

public void updateByHashcode(HllBuffer buffer,
                             long hash)

Update the HLL++ buffer.

merge

public void merge(HllBuffer buffer1,
                  HllBuffer buffer2)

Merge the HLL buffers by iterating through the registers in both buffers and select the maximum number of leading zeros for each register.

estimateBias

public double estimateBias(double e)

Estimate the bias using the raw estimates with their respective biases from the HLL++ appendix. We currently use KNN interpolation to determine the bias (as suggested in the paper).

query

public long query(HllBuffer buffer)

Compute the HyperLogLog estimate.

Variable names in the HLL++ paper match variable names in the code.

trueRsd

public double trueRsd()

The rsd of HLL++ is always equal to or better than the rsd requested. This method returns the rsd this instance actually guarantees.

Returns:

the actual rsd.