com.amlinv.stats

Class MovingAverage

java.lang.Object

com.amlinv.stats.MovingAverage

public class MovingAverage
extends Object

Missing samples are treated as 0 values. Until at least one sample is recorded, however, the result is treated as no data.

Note that adding elements out-of-order is supported, but is tricky and will lead to poor results unless the values are tightly packed. For example, adding values that pre-date the covered time period has no effect, and adding values far into the future compared to the rest of the values will lead to premature devaluation of the entire average until the intermediate values are filled-in. In addition, updates anywhere besides the end of the list run in linear time - i.e. O(n).

CONCURRENCY WARNING: this class is not safe under concurrency; synchronize all accesses as-needed for concurrent use. Created by art on 5/27/15.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
protected class	MovingAverage.SampleData

Constructor Summary

Constructors

Constructor and Description
MovingAverage(long sampleTimePeriod, long maxSlots)

Method Summary

Methods

Modifier and Type	Method and Description
void	add(long timestamp, long value) Add the given value with the specified timestamp to the average.
protected long	calcNumSlot() Calculate the number of effective slots used by the history, including "sparse" slots - i.e.
protected BigDecimal	decayOldValues(long endSlot) Decay old slots given the new ending slot number.
double	getAverage() Return the moving average of the values collected.
double	getFullPeriodAverage() Return the moving average of the values collected while assuming 0 values for any slots not yet collected.
void	put(long timestamp, long value) Put the given value at the timeslot for the given timestamp, replacing the existing value, if any.
protected void	update(long timestamp, long value, boolean addInd) Update the average history and accumulator with the given value at the given timestamp, adding or replacing the value as specified.

Methods inherited from class java.lang.Object

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

Constructor Detail

MovingAverage

public MovingAverage(long sampleTimePeriod,
             long maxSlots)

Method Detail

put

public void put(long timestamp,
       long value)

Put the given value at the timeslot for the given timestamp, replacing the existing value, if any.

Parameters:

timestamp -

value -

add

public void add(long timestamp,
       long value)

Add the given value with the specified timestamp to the average. If data already exists for the same time slot, the value is added and the average updated.

Parameters:

timestamp -

value -

update

protected void update(long timestamp,
          long value,
          boolean addInd)

Update the average history and accumulator with the given value at the given timestamp, adding or replacing the value as specified.

Parameters:

timestamp - timestamp assigned to the value.

value - value to store in the history for the moving average.

addInd - true => add the value to any existing value; false => replace the current value with the new one.

getAverage

public double getAverage()

Return the moving average of the values collected. Early in the moving average, the values will appear strongly weighted since so few samples are included in the average.

Returns:

average of the collected values.

getFullPeriodAverage

public double getFullPeriodAverage()

Return the moving average of the values collected while assuming 0 values for any slots not yet collected. Early in the moving average, the values will appear weakly weighted since so few actual values are included in the average and a large number of 0 values are assumed.

Returns:

average of the collected values and assumed 0 values to fill out the moving average period.

decayOldValues

protected BigDecimal decayOldValues(long endSlot)

Decay old slots given the new ending slot number.

Parameters:

endSlot - the new ending slot number.

Returns:

sum of the values removed from history.

calcNumSlot

protected long calcNumSlot()

Calculate the number of effective slots used by the history, including "sparse" slots - i.e. slots for which no data is actually stored.

Returns:

number of slots to which the accumulated data applies.