io.micrometer.core.instrument.stats.quantile

Class GKQuantiles

java.lang.Object

io.micrometer.core.instrument.stats.quantile.GKQuantiles

All Implemented Interfaces:

Quantiles

public class GKQuantiles
extends java.lang.Object
implements Quantiles

This class is an implementation of the Greenwald-Khanna algorithm for computing epsilon-approximate quantiles of large data sets. In its pure form it is an offline algorithm. But it is used as a black box by many online algorithms for computing epsilon-approximate quantiles on data streams.
Our implementation widely adapts the original idea published by Michael Greenwald and Sanjeev Khanna in their paper "Space-Efficient Online Computation of Quantile Summaries". Contrary to their idea this implementation uses a list rather than a tree structure to maintain the elements.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	GKQuantiles.Builder

Constructor Summary

Constructors

Constructor and Description
GKQuantiles(java.util.Collection<java.lang.Double> monitored, double epsilon) Creates a new GKQuantiles object that computes epsilon-approximate quantiles.

Method Summary

Modifier and Type	Method and Description
java.lang.Double	get(double q) Estimates appropriate quantiles (i.e.
java.lang.Integer	getCount()
java.util.Collection<java.lang.Double>	monitored() Get all monitored quantiles
void	observe(double value) Add a sample
static GKQuantiles.Builder	quantiles(double... quantiles)
void	setEpsilon(double epsilon)
java.lang.String	toString()

Methods inherited from class java.lang.Object

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

Constructor Detail

GKQuantiles

public GKQuantiles(java.util.Collection<java.lang.Double> monitored,
                   double epsilon)

Creates a new GKQuantiles object that computes epsilon-approximate quantiles.

Parameters:

epsilon - The maximum error bound for quantile estimation.

Method Detail

monitored

public java.util.Collection<java.lang.Double> monitored()

Description copied from interface: Quantiles

Get all monitored quantiles

Specified by:

monitored in interface Quantiles

setEpsilon

public void setEpsilon(double epsilon)

observe

public void observe(double value)

Description copied from interface: Quantiles

Add a sample

Specified by:

observe in interface Quantiles

get

public java.lang.Double get(double q)

Estimates appropriate quantiles (i.e. values that holds epsilon accuracy). Note that if the query parameter doesn't lay in [0,1] Double.NaN is returned! The same result will be returned if an empty instance of GK is queried.

Specified by:

get in interface Quantiles

Parameters:

q - a float value

Returns:

an estimated quantile represented by a Double. Will return Double.NaN if phi isn't between 0 and 1 or this instance of GKQuantiles is empty.

getCount

public java.lang.Integer getCount()

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object

quantiles

public static GKQuantiles.Builder quantiles(double... quantiles)