Package com.tibco.patterns.learn.rlink

Class ConfFeature

  • All Implemented Interfaces:
    Confidence
    Direct Known Subclasses:
    ConfQFeature
    public class ConfFeature
extends java.lang.Object
implements Confidence
    Feature based confidence measure. Immutable class.

    This measure is available for model versions RFV6 and higher (ibi Patterns 5.4). It takes into account the the combination of feature scores seen.

    This measure provides the best overall measure of confidence in the prediction. However it is also quite expensive to compute, so may not be appropriate for use with production query loads.

    • Field Summary

      Fields 
      Modifier and Type Field Description
      static double USE_DFT_BLUR_REGION_WIDTH
      Special value to indicate that the server must use default blur region width.

    • Constructor Summary

      Constructors 
      Constructor Description
      ConfFeature()
      Create an instance with default values.
      ConfFeature​(int minDiff, double blurWindow)
      Create an instance with specific parameter values.

    • Field Detail

      • USE_DFT_BLUR_REGION_WIDTH

        public static final double USE_DFT_BLUR_REGION_WIDTH
        Special value to indicate that the server must use default blur region width.
        See Also:
        Constant Field Values

    • Constructor Detail

      • ConfFeature

        public ConfFeature()
        Create an instance with default values.

      • ConfFeature

        public ConfFeature​(int minDiff,
                   double blurWindow)
        Create an instance with specific parameter values.
        Parameters:
        minDiff - The number of unambiguous training examples needed to reach full confidence. This is the number of unambiguous examples that trained the same combination of feature scores: ABS(True examples - False examples). Only the examples seen in the first training iteration are counted.
        blurWindow - When considering the region trained by a particular set of scores a certain amount of leeway, or blurring of the region can be introduced. This accounts for the way model training functions. This option sets the width of the blur region. A value of 0.0 uses no blur, a value of 1.0 blurs the entire area. The default value is 0.18.
        Throws:
        java.lang.IllegalArgumentException - if a parameter is not in legal range.

    • Method Detail

      • toString

        public java.lang.String toString()
        Overrides:
        toString in class java.lang.Object

      • getConfType

        public int getConfType()
        Return the confidence type.
        Specified by:
        getConfType in interface Confidence
        Returns:
        the confidence type for the Feature-based measure.

      • getMinDiff

        public int getMinDiff()
        Returns:
        number of training examples that trained a particular feature score combination needed for full confidence. Always positive.

      • getBlurWindow

        public double getBlurWindow()
        Returns:
        the width of the blur region (negative if default is used).

      • getIntArg

        public int getIntArg​(int idx)
        Return an integer parameter. The first parameter is the minimum difference count.
        Specified by:
        getIntArg in interface Confidence
        Parameters:
        idx - the integer parameter index, zero based.
        Returns:
        The minimum difference count for index 0, -1 otherwise.

      • getDblArg

        public double getDblArg​(int idx)
        Return a double parameter. The blur window size.
        Specified by:
        getDblArg in interface Confidence
        Parameters:
        idx - the double parameter index, zero based.
        Returns:
        index 0: the blur window size. -1.0 for other indexes.