Class LFSMethods

java.lang.Object
weka.attributeSelection.LFSMethods
All Implemented Interfaces:
RevisionHandler

public class LFSMethods extends Object implements RevisionHandler
Version:
$Revision: 1.3 $
Author:
Martin Guetlein (martin.guetlein@gmail.com)
  • Constructor Details

    • LFSMethods

      public LFSMethods()
      empty constructor methods are not static because of access to inner class Link2 and LinkedList2
  • Method Details

    • getBestGroup

      public BitSet getBestGroup()
      Returns:
      best group found by forwardSearch/floatingForwardSearch
    • getBestMerit

      public double getBestMerit()
      Returns:
      merit of best group found by forwardSearch/floatingForwardSearch
    • getBestGroupOfSize

      public BitSet getBestGroupOfSize(int size)
      Returns:
      best group of size found by forwardSearch
    • getNumEvalsCached

      public int getNumEvalsCached()
      Returns:
      number of cached / not performed evaluations
    • getNumEvalsTotal

      public int getNumEvalsTotal()
      Returns:
      number totally performed evaluations
    • rankAttributes

      public int[] rankAttributes(Instances data, SubsetEvaluator evaluator, boolean verbose) throws Exception
      Returns:
      ranking (integer array) of attributes in data with evaluator (sorting is NOT stable!)
      Throws:
      Exception
    • forwardSearch

      public BitSet forwardSearch(int cacheSize, BitSet startGroup, int[] ranking, int k, boolean incrementK, int maxStale, int forceResultSize, Instances data, SubsetEvaluator evaluator, boolean verbose) throws Exception
      Performs linear forward selection
      Parameters:
      cacheSize - chacheSize (times number of instances) to store already evaluated sets
      startGroup - start group for search (can be null)
      ranking - ranking of attributes (as produced by rankAttributes), no ranking would be [0,1,2,3,4..]
      k - number of top k attributes that are taken into account
      incrementK - true -> fixed-set, false -> fixed-width
      maxStale - number of times the search proceeds even though no improvement was found (1 = hill-climbing)
      forceResultSize - stopping criteria changed from no-improvement (forceResultSize=-1) to subset-size
      data -
      evaluator -
      verbose -
      Returns:
      BitSet, that cotains the best-group found
      Throws:
      Exception
    • floatingForwardSearch

      public BitSet floatingForwardSearch(int cacheSize, BitSet startGroup, int[] ranking, int k, boolean incrementK, int maxStale, Instances data, SubsetEvaluator evaluator, boolean verbose) throws Exception
      Performs linear floating forward selection ( the stopping criteria cannot be changed to a specific size value )
      Parameters:
      cacheSize - chacheSize (times number of instances) to store already evaluated sets
      startGroup - start group for search (can be null)
      ranking - ranking of attributes (as produced by rankAttributes), no ranking would be [0,1,2,3,4..]
      k - number of top k attributes that are taken into account
      incrementK - true -> fixed-set, false -> fixed-width
      maxStale - number of times the search proceeds even though no improvement was found (1 = hill-climbing)
      data -
      evaluator -
      verbose -
      Returns:
      BitSet, that cotains the best-group found
      Throws:
      Exception
    • getRevision

      public String getRevision()
      Returns the revision string.
      Specified by:
      getRevision in interface RevisionHandler
      Returns:
      the revision