com.pervasive.datarush.analytics.knn

Class NearnessMeasure

java.lang.Object

com.pervasive.datarush.analytics.knn.NearnessMeasure

All Implemented Interfaces:

Serializable

public abstract class NearnessMeasure
extends Object
implements Serializable

Determines how to measure how near an example and query record are to each other.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static interface	NearnessMeasure.Measurer A calculator of "nearness".

Constructor Summary

Constructors

Constructor and Description
NearnessMeasure()

Method Summary

Methods

Modifier and Type	Method and Description
static NearnessMeasure	cosine() Measures nearness using cosine similarity.
static NearnessMeasure	euclidean() Measures nearness using Euclidean distance.
abstract double	nearnessToDistance(double n) Converts a "nearness" value to a "real" distance.
abstract NearnessMeasure.Measurer	newMeasurer() Creates a new measurer for use.

Methods inherited from class java.lang.Object

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

Constructor Detail

NearnessMeasure

public NearnessMeasure()

Method Detail

newMeasurer

public abstract NearnessMeasure.Measurer newMeasurer()

Creates a new measurer for use.

Returns:

a new measure calculator

nearnessToDistance

public abstract double nearnessToDistance(double n)

Converts a "nearness" value to a "real" distance. This allows avoidance of expensive computations in determining how near two points are to one another.

For example, in Euclidean distance, a square root is required to obtain the actual distance. However, the squared distance provides exactly the same ordering as the true distance, so can be used as the "nearness". If the true distance is required, it can be derived on an as needed basis.

Parameters:

n - the "nearness" value returned by a measurer

Returns:

the actual distance between the points

cosine

public static NearnessMeasure cosine()

Measures nearness using cosine similarity.

Results in the range [0,-1] have no clear interpretation of "nearness". In this implementation, 0 is arbitrarily considered "nearer" than -1. Due to this ambiguity, this measure is best used when the values of features are non-negative, ensuring results in the range [1,0] for which the interpretation of "nearer" is clear.

Returns:

a measurement based on cosine similarity

euclidean

public static NearnessMeasure euclidean()

Measures nearness using Euclidean distance.

Returns:

a measurement based on Euclidean distance