Peano

Searching...

No Matches

#include <Measurement.h>

Collaboration diagram for tarch::timing::Measurement:

[legend]

Public Member Functions
	Measurement (const double &accuracy=0.0)
	~Measurement ()=default
double	getValue () const
double	getMostRecentValue () const
double	getAccumulatedValue () const
double	getStandardDeviation () const
	We did face some seg faults (very rarely) where the parameter under the square root did become slightly negative.
bool	isAccurateValue () const
	Is value accurate.
void	setAccuracy (const double &value)
void	increaseAccuracy (const double &factor)
void	setValue (const double &value)
	Set the value.
int	getNumberOfMeasurements () const
std::string	toString () const
double	max () const
double	min () const
void	erase ()
double	getAccuracy () const

Private Member Functions
double	getMeanValue () const
double	getMeanValueOfNextStep (double newValue) const

Private Attributes
double	_accuracy
double	_accumulatedValue
	Accumulated value.
double	_lastValue
	We store the last value.
double	_accumulatedSquares
	To compute the standard deviation, we rely on the formula.
double	_numberOfMeasurements
	Needed to compute average value and variance.
bool	_isAccurateValue
double	_min
double	_max
double	_minMeasurement
double	_maxMeasurement

Static Private Attributes
static tarch::logging::Log	_log

Detailed Description

Measurement.

The shared memory oracles have to handle real time measurements. This data is always noisy, inaccurate, and has to be averaged. We originally planned to move all the averaging etc. to the oracle singleton. However, that is hardly possible, as the responsible oracle changes all the time, and the measurements are associated to one oracle. Consequently, we introduced this helper class for the individual oracle implementations.

Author: Tobias Weinzierl

Definition at line 29 of file Measurement.h.

Constructor & Destructor Documentation

◆ Measurement()

tarch::timing::Measurement::Measurement ( const double & accuracy = 0.0 )

Definition at line 18 of file Measurement.cpp.

References _accumulatedSquares, _accumulatedValue, _accuracy, _isAccurateValue, _lastValue, _max, _maxMeasurement, _min, _minMeasurement, _numberOfMeasurements, assertion1, double, and max().

Here is the call graph for this function:

◆ ~Measurement()

tarch::timing::Measurement::~Measurement ( )

default

Member Function Documentation

◆ erase()

void tarch::timing::Measurement::erase ( )

Definition at line 34 of file Measurement.cpp.

References _accumulatedSquares, _accumulatedValue, _isAccurateValue, _max, _maxMeasurement, _min, _minMeasurement, and _numberOfMeasurements.

◆ getAccumulatedValue()

double tarch::timing::Measurement::getAccumulatedValue ( ) const

Returns: Sum over all values measured so far.

Definition at line 58 of file Measurement.cpp.

References _accumulatedValue.

◆ getAccuracy()

double tarch::timing::Measurement::getAccuracy ( ) const

Definition at line 186 of file Measurement.cpp.

References _accuracy.

◆ getMeanValue()

double tarch::timing::Measurement::getMeanValue ( ) const

private

Definition at line 46 of file Measurement.cpp.

References _accumulatedValue, _numberOfMeasurements, and assertion.

Referenced by getStandardDeviation(), getValue(), and setValue().

Here is the caller graph for this function:

◆ getMeanValueOfNextStep()

double tarch::timing::Measurement::getMeanValueOfNextStep ( double newValue ) const

private

Definition at line 52 of file Measurement.cpp.

References _accumulatedValue, _numberOfMeasurements, and assertion.

Referenced by setValue().

Here is the caller graph for this function:

◆ getMostRecentValue()

double tarch::timing::Measurement::getMostRecentValue ( ) const

Returns: Last value added. If there's none so far, the routine returns 0. If you've only added one value so far, the result equals getValue().

Definition at line 61 of file Measurement.cpp.

References _lastValue.

◆ getNumberOfMeasurements()

int tarch::timing::Measurement::getNumberOfMeasurements ( ) const

Definition at line 136 of file Measurement.cpp.

References _numberOfMeasurements.

◆ getStandardDeviation()

double tarch::timing::Measurement::getStandardDeviation ( ) const

We did face some seg faults (very rarely) where the parameter under the square root did become slightly negative.

This seems to be a round-off error. We circumnavigate it by adding an absolute value which is mathematically not required.

Definition at line 76 of file Measurement.cpp.

References _accumulatedSquares, _accumulatedValue, _numberOfMeasurements, std::abs(), assertionEquals, and getMeanValue().

Referenced by toString().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getValue()

double tarch::timing::Measurement::getValue ( ) const

Returns: Averaged value (mean value) of all measurements. If you have not yet added a value, it returns 0.

Definition at line 66 of file Measurement.cpp.

References _accumulatedValue, _numberOfMeasurements, assertionEquals, and getMeanValue().

Referenced by assessKernel(), reportRuntime(), reportRuntime(), and toString().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ increaseAccuracy()

void tarch::timing::Measurement::increaseAccuracy ( const double & factor )

Definition at line 95 of file Measurement.cpp.

References _accuracy, and assertion.

◆ isAccurateValue()

bool tarch::timing::Measurement::isAccurateValue ( ) const

Is value accurate.

Whether a value is accurate depends on the last setValue() call. The class internally holds the mean value of all setValue() calls. If a new value is set/added, the object checks whether this additional measurement modifies the mean value more than the given accuracy. If this is the case, the object does not represent a valid measurement yet.

It does not work if we just compare new values to the mean value. If a measurement looks similar to the sketch from above (black=mean value, grey area=accuracy, blue points=values), the measurement never would become valid.

I therefore use a normalised mean value

\( valid = | \frac{ m^{(n+1)} - m^{(n)} }{m^{(n)}} | \leq \epsilon = | \frac{m^{(n+1)}}{m^{(n)}}-1 | \leq \epsilon \)

where \( m^{(n)} \) and \( m^{(n+1)} \) are two subsequent mean values.

Definition at line 86 of file Measurement.cpp.

References _isAccurateValue.

Referenced by max(), and min().

Here is the caller graph for this function: