org.dllearner.algorithms.refinement2
Class MultiHeuristic

java.lang.Object
  org.dllearner.algorithms.refinement2.MultiHeuristic

All Implemented Interfaces:

Comparator<ExampleBasedNode>, ExampleBasedHeuristic

public class MultiHeuristic
extends Object
implements ExampleBasedHeuristic

This heuristic combines the following criteria to assign a double score value to a node:

• quality/accuracy of a concept (based on the full training set, not the negative example coverage as the flexible heuristic)
• horizontal expansion
• accuracy gain: The heuristic takes into account the accuracy difference between a node and its parent. If there is no gain (even though we know that the refinement is proper) it is unlikely (although not excluded) that the refinement is a necessary path to take towards a solution.

The heuristic has two parameters:

• expansion penalty factor: describes how much accuracy gain is worth an increase of horizontal expansion by one (typical value: 0.01)
• gain bonus factor: describes how accuracy gain should be weighted versus accuracy itself (typical value: 1.00)

The value of a node is calculated as follows:

value = accuracy + gain bonus factor * accuracy gain - expansion penalty factor * horizontal expansion - node children penalty factor * number of children of node

accuracy = (TP + TN)/(P + N)

TP = number of true positives (= covered positives)
TN = number of true negatives (= nr of negatives examples - covered negatives)
P = number of positive examples
N = number of negative examples


Author:

Jens Lehmann

Constructor Summary

MultiHeuristic(int nrOfPositiveExamples, int nrOfNegativeExamples)
Deprecated.

MultiHeuristic(int nrOfPositiveExamples, int nrOfNegativeExamples, ROLComponent2Configurator configurator)

Method Summary

int	compare(ExampleBasedNode node1, ExampleBasedNode node2)
double	getNodeScore(ExampleBasedNode node)
static double	getNodeScore(ExampleBasedNode node, int nrOfPositiveExamples, int nrOfNegativeExamples, ROLComponent2Configurator configurator)

Methods inherited from class java.lang.Object

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

Methods inherited from interface java.util.Comparator

equals

Constructor Detail

MultiHeuristic

@Deprecated
public MultiHeuristic(int nrOfPositiveExamples,
                                 int nrOfNegativeExamples)

Deprecated.

MultiHeuristic

public MultiHeuristic(int nrOfPositiveExamples,
                      int nrOfNegativeExamples,
                      ROLComponent2Configurator configurator)

Method Detail

compare

public int compare(ExampleBasedNode node1,
                   ExampleBasedNode node2)

Specified by:

compare in interface Comparator<ExampleBasedNode>

getNodeScore

public double getNodeScore(ExampleBasedNode node)

getNodeScore

public static double getNodeScore(ExampleBasedNode node,
                                  int nrOfPositiveExamples,
                                  int nrOfNegativeExamples,
                                  ROLComponent2Configurator configurator)