tv.danmaku.ijk.media.exo

Class SlidingPercentile

java.lang.Object

tv.danmaku.ijk.media.exo.SlidingPercentile

public class SlidingPercentile
extends java.lang.Object

Calculate any percentile over a sliding window of weighted values. A maximum weight is configured. Once the total weight of the values reaches the maximum weight, the oldest value is reduced in weight until it reaches zero and is removed. This maintains a constant total weight, equal to the maximum allowed, at the steady state.

This class can be used for bandwidth estimation based on a sliding window of past transfer rate observations. This is an alternative to sliding mean and exponential averaging which suffer from susceptibility to outliers and slow adaptation to step functions.

See Also:

Wiki: Moving average, Wiki: Selection algorithm

Constructor Summary

Constructors

Constructor and Description
SlidingPercentile(int maxWeight)

Method Summary

Modifier and Type	Method and Description
void	addSample(int weight, float value) Adds a new weighted value.
float	getPercentile(float percentile) Computes a percentile by integration.
void	reset() Resets the sliding percentile.

Methods inherited from class java.lang.Object

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

Constructor Detail

SlidingPercentile

public SlidingPercentile(int maxWeight)

Parameters:

maxWeight - The maximum weight.

Method Detail

reset

public void reset()

Resets the sliding percentile.

addSample

public void addSample(int weight,
                      float value)

Adds a new weighted value.

Parameters:

weight - The weight of the new observation.

value - The value of the new observation.

getPercentile

public float getPercentile(float percentile)

Computes a percentile by integration.

Parameters:

percentile - The desired percentile, expressed as a fraction in the range (0,1].

Returns:

The requested percentile value or Float.NaN if no samples have been added.