de.unibi.techfak.jpredictor.motifs
Class MotifEvolutionRewardedSelection

java.lang.Object
  de.unibi.techfak.jpredictor.motifs.MotifEvolution
      de.unibi.techfak.jpredictor.motifs.MotifEvolutionES
          de.unibi.techfak.jpredictor.motifs.MotifEvolutionRewardedSelection

public class MotifEvolutionRewardedSelection
extends MotifEvolutionES

Motif evolution with a reward system for selecting motifs. The reward value is measured in terms of entropy (p log p) and multiplied to the weight.

Field Summary

(package private) java.util.Vector[]

posOcc For every motif one vector, which contains the number of occurrences in the positive training set.

Fields inherited from class de.unibi.techfak.jpredictor.motifs.MotifEvolution

comm, generatingCharacters, generatingDistribution, gML, motifCounter, op, temperature

Constructor Summary

MotifEvolutionRewardedSelection(ICommunicator comm)

Method Summary

(package private) void	fillNewWeights(MotifList ml, double[] w) This method fills the array w with the new weights for the motifs.
(package private) boolean	weightMotifLists() Only weights the child motif list.

Methods inherited from class de.unibi.techfak.jpredictor.motifs.MotifEvolutionES

evolveMotifLists, initMotifLists, selectMotifLists

Methods inherited from class de.unibi.techfak.jpredictor.motifs.MotifEvolution

evolveMultiMotif, outputResults, recombineParentSet, restrainMotifList, start, weightMotifs

Methods inherited from class java.lang.Object

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

Field Detail

posOcc

java.util.Vector[] posOcc

For every motif one vector, which contains the number of occurrences in the positive training set. The array is filled by a call to weightMotifLists() in this class and used by the fillNewWeights(MotifList, double[])-method for changing the motifs weights before the list is restrained.

See Also:

weightMotifLists(), MotifEvolution.weightMotifs(Vector[], Vector[]), fillNewWeights(MotifList, double[])

Constructor Detail

MotifEvolutionRewardedSelection

public MotifEvolutionRewardedSelection(ICommunicator comm)
                                throws java.lang.NullPointerException,
                                       java.lang.IllegalArgumentException

Throws:

java.lang.NullPointerException

java.lang.IllegalArgumentException

Method Detail

weightMotifLists

boolean weightMotifLists()

Only weights the child motif list.

Overrides:

weightMotifLists in class MotifEvolution

Returns:

true, if weighting was successful, false otherwise.

See Also:

MotifEvolution.start(int), MotifEvolution.weightMotifs(Vector[], Vector[])

fillNewWeights

void fillNewWeights(MotifList ml,
                    double[] w)

This method fills the array w with the new weights for the motifs. Overwrites the super method to implement the reward system. The new weights are taken the product of the old ones times the entropy, which is the sum of (p log p), where p is the relative occurrence of the motif in the positive training set.

Overrides:

fillNewWeights in class MotifEvolution

Parameters:

ml - The list of weighted motifs.

w - An array at least the size of the motif list to be filled with the new weights for the motifs.