Sleipnir::IMeasure Class Reference

Encapsulates any similarity (or occasionally distance) measure operating over two vectors. More...

#include <measure.h>

Inheritance diagram for Sleipnir::IMeasure:

Public Types
enum	EMap { EMapNone = 0, EMapCenter = EMapNone + 1, EMapAbs = EMapCenter + 1 }
	Indicates how the result of a measure should be centered. More...
Public Member Functions
virtual const char *	GetName () const =0
	Return the human-readable unique identifier of the measure type.
virtual bool	IsRank () const =0
	Return true if the measure requires rank-based integer inputs.
virtual IMeasure *	Clone () const =0
	Create a copy of the current measure object.
virtual double	Measure (const float *adX, size_t iN, const float *adY, size_t iM, EMap eMap=EMapNone, const float *adWX=NULL, const float *adWY=NULL) const =0
	Calculate the measure between two given vectors with optional element weights.

---

Detailed Description

Encapsulates any similarity (or occasionally distance) measure operating over two vectors.

IMeasures are generally similarity measures consuming two floating point vectors and producing a single continuous result. There are a few distance measures implemented, but since everything in Sleipnir assumes measures to be similarity-based (i.e. higher is closer), several IMeasures exist to reverse distance measures (e.g. CMeasureNegate and CMeasureInvert).

Most measures require input vectors of the same length, but some don't; likewise, most measures will ignore weights for the vector elements if they're provided, but some won't. Most measures will deal appropriately with missing values (NaNs) in their inputs. Many measures support different forms of centering, although some will ignore centering and return the raw measure value. For more information, see EMap.

Remarks:

Most measure objects don't actually use any memory, but a few do, depending on their implementation, and care should be taken to manage memory for these appropriately. The Clone method exists to allow a measure to copy itself without breaking the interface, and most measures that require memory management take a constructor argument indicating whether they are responsible for freeing memory references.

Definition at line 52 of file measure.h.

---

Member Enumeration Documentation

enum Sleipnir::IMeasure::EMap

Indicates how the result of a measure should be centered.

Enumerator:

EMapNone	Perform no centering and return the original measure result.
EMapCenter	Center the measure result by scaling to the range [0, 1].
EMapAbs	Return the absolute value of the original measure result.

Definition at line 58 of file measure.h.

---

Member Function Documentation

virtual IMeasure* Sleipnir::IMeasure::Clone

(

)

const [pure virtual]

Create a copy of the current measure object.

Returns:

Copy of the current measure object.

Remarks:

Caller is, of course, responsible for destroying the created object.

Implemented in Sleipnir::CMeasureDice, Sleipnir::CMeasureSignedDistanceCorrelation, Sleipnir::CMeasureDistanceCorrelation, Sleipnir::CMeasurePearsonSignificance, Sleipnir::CMeasureRelativeAUC, Sleipnir::CMeasureMutualInformation, Sleipnir::CMeasureBinaryInnerProduct, Sleipnir::CMeasureInnerProduct, Sleipnir::CMeasureHypergeometric, Sleipnir::CMeasurePearNorm, Sleipnir::CMeasureSpearman, Sleipnir::CMeasureKendallsTau, Sleipnir::CMeasureKolmogorovSmirnov, Sleipnir::CMeasureQuickPearson, Sleipnir::CMeasurePearson, Sleipnir::CMeasureEuclideanScaled, Sleipnir::CMeasureEuclidean, Sleipnir::CMeasureAutocorrelate, Sleipnir::CMeasureInvert, Sleipnir::CMeasureNegate, and Sleipnir::CMeasureSigmoid.

Referenced by Sleipnir::CMeasureSigmoid::Clone(), Sleipnir::CMeasureNegate::Clone(), Sleipnir::CMeasureInvert::Clone(), Sleipnir::CMeasureAutocorrelate::Clone(), and Sleipnir::CPCL::Distance().

virtual const char* Sleipnir::IMeasure::GetName

(

)

const [pure virtual]

Return the human-readable unique identifier of the measure type.

Returns:

The string identifier of the measure type.

Implemented in Sleipnir::CMeasureDice, Sleipnir::CMeasureSignedDistanceCorrelation, Sleipnir::CMeasureDistanceCorrelation, Sleipnir::CMeasurePearsonSignificance, Sleipnir::CMeasureRelativeAUC, Sleipnir::CMeasureMutualInformation, Sleipnir::CMeasureBinaryInnerProduct, Sleipnir::CMeasureInnerProduct, Sleipnir::CMeasureHypergeometric, Sleipnir::CMeasurePearNorm, Sleipnir::CMeasureSpearman, Sleipnir::CMeasureKendallsTau, Sleipnir::CMeasureKolmogorovSmirnov, Sleipnir::CMeasureQuickPearson, Sleipnir::CMeasurePearson, Sleipnir::CMeasureEuclideanScaled, Sleipnir::CMeasureEuclidean, Sleipnir::CMeasureAutocorrelate, Sleipnir::CMeasureInvert, Sleipnir::CMeasureNegate, and Sleipnir::CMeasureSigmoid.

Referenced by Sleipnir::CPCL::Distance(), Sleipnir::CMeasureSigmoid::GetName(), Sleipnir::CMeasureNegate::GetName(), Sleipnir::CMeasureInvert::GetName(), and Sleipnir::CMeasureAutocorrelate::GetName().

virtual bool Sleipnir::IMeasure::IsRank

(

)

const [pure virtual]

Return true if the measure requires rank-based integer inputs.

Returns:

True if the measure requires rank-based input vectors.

Remarks:

The input vectors for rank measures should contain only floating point values with no fractional part; behavior is undefined if they don't.

Implemented in Sleipnir::CMeasureDice, Sleipnir::CMeasureSignedDistanceCorrelation, Sleipnir::CMeasureDistanceCorrelation, Sleipnir::CMeasurePearsonSignificance, Sleipnir::CMeasureRelativeAUC, Sleipnir::CMeasureMutualInformation, Sleipnir::CMeasureBinaryInnerProduct, Sleipnir::CMeasureInnerProduct, Sleipnir::CMeasureHypergeometric, Sleipnir::CMeasurePearNorm, Sleipnir::CMeasureSpearman, Sleipnir::CMeasureKendallsTau, Sleipnir::CMeasureKolmogorovSmirnov, Sleipnir::CMeasureQuickPearson, Sleipnir::CMeasurePearson, Sleipnir::CMeasureEuclideanScaled, Sleipnir::CMeasureEuclidean, Sleipnir::CMeasureAutocorrelate, Sleipnir::CMeasureInvert, Sleipnir::CMeasureNegate, and Sleipnir::CMeasureSigmoid.

Referenced by Sleipnir::CPCL::Distance(), Sleipnir::CMeasureSigmoid::IsRank(), Sleipnir::CMeasureNegate::IsRank(), Sleipnir::CMeasureInvert::IsRank(), Sleipnir::CMeasureAutocorrelate::IsRank(), and Sleipnir::CDatasetCompact::Open().

virtual double Sleipnir::IMeasure::Measure	(	const float *	adX,
		size_t	iN,
		const float *	adY,
		size_t	iM,
		EMap	eMap = EMapNone,
		const float *	adWX = NULL,
		const float *	adWY = NULL
	)		const [pure virtual]

Calculate the measure between two given vectors with optional element weights.

Parameters:

adX	First array of values.
iN	Length of first array.
adY	Second array of values.
iM	Length of second array.
eMap	Way in which returned value should be centered (implementation-specific).
adWX	If non-null, weights of elements in the first array.
adWY	If non-null, weights of elements in the second array.

Returns:

Measure calculated between the two input vectors and, optionally, weights.

Remarks:

Pretty much every implementation will puke if given bad input; bounds checking etc. is minimal.

Implemented in Sleipnir::CMeasureDice, Sleipnir::CMeasureSignedDistanceCorrelation, Sleipnir::CMeasureDistanceCorrelation, Sleipnir::CMeasurePearsonSignificance, Sleipnir::CMeasureRelativeAUC, Sleipnir::CMeasureMutualInformation, Sleipnir::CMeasureBinaryInnerProduct, Sleipnir::CMeasureInnerProduct, Sleipnir::CMeasureHypergeometric, Sleipnir::CMeasurePearNorm, Sleipnir::CMeasureSpearman, Sleipnir::CMeasureKendallsTau, Sleipnir::CMeasureKolmogorovSmirnov, Sleipnir::CMeasureQuickPearson, Sleipnir::CMeasurePearson, Sleipnir::CMeasureEuclideanScaled, Sleipnir::CMeasureEuclidean, Sleipnir::CMeasureAutocorrelate, Sleipnir::CMeasureInvert, Sleipnir::CMeasureNegate, and Sleipnir::CMeasureSigmoid.

Referenced by Sleipnir::CClustKMeans::Cluster(), Sleipnir::CPCL::Distance(), Sleipnir::CPCL::Impute(), Sleipnir::CMeasureSigmoid::Measure(), Sleipnir::CMeasureNegate::Measure(), Sleipnir::CMeasureInvert::Measure(), Sleipnir::CMeasureAutocorrelate::Measure(), and Sleipnir::CDatasetCompact::Open().

---

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

• src/measure.h