DirectEWMAX< P > Class Template Reference

An exponentially weighted moving average. More...

#include <click/ewma.hh>

Inherits P.

Public Types
typedef P::value_type	value_type

Public Member Functions
	DirectEWMAX ()
	Construct a EWMA with initial average 0.
	DirectEWMAX (value_type scaled_value)
	Construct a EWMA with initial scaled average scaled_value.
value_type	scaled_average () const
	Return the current scaled moving average.
value_type	unscaled_average () const
	Return the current moving average, rounded up.
void	clear ()
	Reset the EWMA to value 0.
void	assign (value_type scaled_value)
	Assign the EWMA to scaled average scaled_value.
void	update (value_type x)
	Update the moving average with a new observation.
void	update_n (value_type x, unsigned n)
	Update the moving average with n identical observations.
String	unparse () const
	Unparse the current average into a String.
void	update_with (value_type x)
	Update the moving average with a new observation (deprecated).

Detailed Description

template<typename P>
class DirectEWMAX< P >

An exponentially weighted moving average.

The DirectEWMAX template class represents a simple exponentially weighted moving average. The average starts out with value 0. The update() function adds a new observation to the average.

The template parameter P defines three EWMA parameters: value type, stability shift, and scale factor.

The value type is simply the unsigned integral type used to store the average. It is also the type of each observation. unsigned and uint64_t are examples.

The stability shift specifies alpha, the stability parameter. Concretely, alpha = 1. / (2 ** stability_shift). Thus, a stability shift of 4 corresponds to an alpha of 1/16.

The scaling factor specifies how many bits of fraction are kept per observation. Fraction bits are necessary to account for small differences between observations. For example, consider a EWMA with value 0, alpha 1/16, and 0 bits of fraction. Assume the EWMA begins to observe a stream of observations equal to 1. Despite these observations, the EWMA's value will never change from 0, since the 1/16 alpha factor rounds the new observations down to 0. At least 4 bits of fraction are required to account for this difference. There is a tradeoff: the more bits of fraction, the more precise the EWMA, but the less bits available to account for large values.

These EWMA parameters are defined by five of P's members, two typedefs and three possibly static member functions.

P::value_type

The EWMA's value type. Example: unsigned.

P::signed_value_type

The signed version of P::value_type. Used internally. Example: int.

unsigned P::stability_shift()

This function should return this EWMA's stability shift (see above).

unsigned P::scale()

This function should return this EWMA's scaling factor (see above).

unsigned P::compensation()

This function should return this EWMA's stability compensation, which normally equals 1 << (stability_shift - 1).

Since DirectEWMAX inherits from an object of type P, these members are also directly available to callers.

The FixedEWMAXParameters and StabilityEWMAXParameters types are good template arguments for DirectEWMAX.

See also

RateEWMAX

Constructor & Destructor Documentation

DirectEWMAX() [1/2]

template<typename P >

DirectEWMAX< P >::DirectEWMAX ( )

inline

Construct a EWMA with initial average 0.

DirectEWMAX() [2/2]

template<typename P >

DirectEWMAX< P >::DirectEWMAX ( value_type  scaled_value )

inline

Construct a EWMA with initial scaled average scaled_value.

Member Function Documentation

scaled_average()

template<typename P >

value_type DirectEWMAX< P >::scaled_average ( ) const

inline

Return the current scaled moving average.

Note

The returned value has scale() bits of fraction.

unscaled_average()

template<typename P >

value_type DirectEWMAX< P >::unscaled_average ( ) const

inline

Return the current moving average, rounded up.

Note

The returned value is unscaled (has zero bits of fraction).

clear()

template<typename P >

void DirectEWMAX< P >::clear ( )

inline

Reset the EWMA to value 0.

assign()

template<typename P >

void DirectEWMAX< P >::assign ( value_type  scaled_value )

inline

Assign the EWMA to scaled average scaled_value.

update()

template<typename P >

void DirectEWMAX< P >::update ( value_type  x )

inline

Update the moving average with a new observation.

Parameters

x	the observation (unscaled)

update_n()

template<typename P >

void DirectEWMAX< P >::update_n	(	value_type	x,
		unsigned	n
	)

Update the moving average with n identical observations.

Parameters

x	the observation (unscaled)
n	number of observations

Note

This may be faster than calling update(x) n times.

unparse()

template<typename P >

String DirectEWMAX< P >::unparse ( ) const

inline

Unparse the current average into a String.

Note

The returned value is unscaled, but may contain a fractional part.

update_with()

template<typename P >

void DirectEWMAX< P >::update_with ( value_type  x )

inline

Update the moving average with a new observation (deprecated).

Parameters

x	the observation (unscaled)

Deprecated:

Use update() instead.

---

The documentation for this class was generated from the following file:

• include/click/ewma.hh