Module datarush.analytics
Package 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:

  • Constructor Details

    • NearnessMeasure

      public NearnessMeasure()

  • Method Details

    • 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