/*
* This program is free software; you can redistribsute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
/*
* DataNearBalancedND.java
* Copyright (C) 2005 University of Waikato, Hamilton, New Zealand
*
*/
package weka.classifiers.meta.nestedDichotomies;
import weka.classifiers.Classifier;
import weka.classifiers.RandomizableSingleClassifierEnhancer;
import weka.classifiers.meta.FilteredClassifier;
import weka.core.Capabilities;
import weka.core.Instance;
import weka.core.Instances;
import weka.core.Range;
import weka.core.RevisionUtils;
import weka.core.TechnicalInformation;
import weka.core.TechnicalInformationHandler;
import weka.core.Utils;
import weka.core.Capabilities.Capability;
import weka.core.TechnicalInformation.Field;
import weka.core.TechnicalInformation.Type;
import weka.filters.Filter;
import weka.filters.unsupervised.attribute.MakeIndicator;
import weka.filters.unsupervised.instance.RemoveWithValues;
import java.util.Hashtable;
import java.util.Random;
import weka.classifiers.AbstractClassifier;
/**
<!-- globalinfo-start -->
* A meta classifier for handling multi-class datasets with 2-class classifiers by building a random data-balanced tree structure.<br/>
* <br/>
* For more info, check<br/>
* <br/>
* Lin Dong, Eibe Frank, Stefan Kramer: Ensembles of Balanced Nested Dichotomies for Multi-class Problems. In: PKDD, 84-95, 2005.<br/>
* <br/>
* Eibe Frank, Stefan Kramer: Ensembles of nested dichotomies for multi-class problems. In: Twenty-first International Conference on Machine Learning, 2004.
* <p/>
<!-- globalinfo-end -->
*
<!-- technical-bibtex-start -->
* BibTeX:
* <pre>
* @inproceedings{Dong2005,
* author = {Lin Dong and Eibe Frank and Stefan Kramer},
* booktitle = {PKDD},
* pages = {84-95},
* publisher = {Springer},
* title = {Ensembles of Balanced Nested Dichotomies for Multi-class Problems},
* year = {2005}
* }
*
* @inproceedings{Frank2004,
* author = {Eibe Frank and Stefan Kramer},
* booktitle = {Twenty-first International Conference on Machine Learning},
* publisher = {ACM},
* title = {Ensembles of nested dichotomies for multi-class problems},
* year = {2004}
* }
* </pre>
* <p/>
<!-- technical-bibtex-end -->
*
<!-- options-start -->
* Valid options are: <p/>
*
* <pre> -S <num>
* Random number seed.
* (default 1)</pre>
*
* <pre> -D
* If set, classifier is run in debug mode and
* may output additional info to the console</pre>
*
* <pre> -W
* Full name of base classifier.
* (default: weka.classifiers.trees.J48)</pre>
*
* <pre>
* Options specific to classifier weka.classifiers.trees.J48:
* </pre>
*
* <pre> -U
* Use unpruned tree.</pre>
*
* <pre> -C <pruning confidence>
* Set confidence threshold for pruning.
* (default 0.25)</pre>
*
* <pre> -M <minimum number of instances>
* Set minimum number of instances per leaf.
* (default 2)</pre>
*
* <pre> -R
* Use reduced error pruning.</pre>
*
* <pre> -N <number of folds>
* Set number of folds for reduced error
* pruning. One fold is used as pruning set.
* (default 3)</pre>
*
* <pre> -B
* Use binary splits only.</pre>
*
* <pre> -S
* Don't perform subtree raising.</pre>
*
* <pre> -L
* Do not clean up after the tree has been built.</pre>
*
* <pre> -A
* Laplace smoothing for predicted probabilities.</pre>
*
* <pre> -Q <seed>
* Seed for random data shuffling (default 1).</pre>
*
<!-- options-end -->
*
* @author Lin Dong
* @author Eibe Frank
*/
public class DataNearBalancedND
extends RandomizableSingleClassifierEnhancer
implements TechnicalInformationHandler {
/** for serialization */
static final long serialVersionUID = 5117477294209496368L;
/** The filtered classifier in which the base classifier is wrapped. */
protected FilteredClassifier m_FilteredClassifier;
/** The hashtable for this node. */
protected Hashtable m_classifiers=new Hashtable();
/** The first successor */
protected DataNearBalancedND m_FirstSuccessor = null;
/** The second successor */
protected DataNearBalancedND m_SecondSuccessor = null;
/** The classes that are grouped together at the current node */
protected Range m_Range = null;
/** Is Hashtable given from END? */
protected boolean m_hashtablegiven = false;
/**
* Constructor.
*/
public DataNearBalancedND() {
m_Classifier = new weka.classifiers.trees.J48();
}
/**
* String describing default classifier.
*
* @return the default classifier classname
*/
protected String defaultClassifierString() {
return "weka.classifiers.trees.J48";
}
/**
* Returns an instance of a TechnicalInformation object, containing
* detailed information about the technical background of this class,
* e.g., paper reference or book this class is based on.
*
* @return the technical information about this class
*/
public TechnicalInformation getTechnicalInformation() {
TechnicalInformation result;
TechnicalInformation additional;
result = new TechnicalInformation(Type.INPROCEEDINGS);
result.setValue(Field.AUTHOR, "Lin Dong and Eibe Frank and Stefan Kramer");
result.setValue(Field.TITLE, "Ensembles of Balanced Nested Dichotomies for Multi-class Problems");
result.setValue(Field.BOOKTITLE, "PKDD");
result.setValue(Field.YEAR, "2005");
result.setValue(Field.PAGES, "84-95");
result.setValue(Field.PUBLISHER, "Springer");
additional = result.add(Type.INPROCEEDINGS);
additional.setValue(Field.AUTHOR, "Eibe Frank and Stefan Kramer");
additional.setValue(Field.TITLE, "Ensembles of nested dichotomies for multi-class problems");
additional.setValue(Field.BOOKTITLE, "Twenty-first International Conference on Machine Learning");
additional.setValue(Field.YEAR, "2004");
additional.setValue(Field.PUBLISHER, "ACM");
return result;
}
/**
* Set hashtable from END.
*
* @param table the hashtable to use
*/
public void setHashtable(Hashtable table) {
m_hashtablegiven = true;
m_classifiers = table;
}
/**
* Generates a classifier for the current node and proceeds recursively.
*
* @param data contains the (multi-class) instances
* @param classes contains the indices of the classes that are present
* @param rand the random number generator to use
* @param classifier the classifier to use
* @param table the Hashtable to use
* @param instsNumAllClasses
* @throws Exception if anything goes worng
*/
private void generateClassifierForNode(Instances data, Range classes,
Random rand, Classifier classifier, Hashtable table,
double[] instsNumAllClasses)
throws Exception {
// Get the indices
int[] indices = classes.getSelection();
// Randomize the order of the indices
for (int j = indices.length - 1; j > 0; j--) {
int randPos = rand.nextInt(j + 1);
int temp = indices[randPos];
indices[randPos] = indices[j];
indices[j] = temp;
}
// Pick the classes for the current split
double total = 0;
for (int j = 0; j < indices.length; j++) {
total += instsNumAllClasses[indices[j]];
}
double halfOfTotal = total / 2;
// Go through the list of classes until the either the left or
// right subset exceeds half the total weight
double sumLeft = 0, sumRight = 0;
int i = 0, j = indices.length - 1;
do {
if (i == j) {
if (rand.nextBoolean()) {
sumLeft += instsNumAllClasses[indices[i++]];
} else {
sumRight += instsNumAllClasses[indices[j--]];
}
} else {
sumLeft += instsNumAllClasses[indices[i++]];
sumRight += instsNumAllClasses[indices[j--]];
}
} while (Utils.sm(sumLeft, halfOfTotal) && Utils.sm(sumRight, halfOfTotal));
int first = 0, second = 0;
if (!Utils.sm(sumLeft, halfOfTotal)) {
first = i;
} else {
first = j + 1;
}
second = indices.length - first;
int[] firstInds = new int[first];
int[] secondInds = new int[second];
System.arraycopy(indices, 0, firstInds, 0, first);
System.arraycopy(indices, first, secondInds, 0, second);
// Sort the indices (important for hash key)!
int[] sortedFirst = Utils.sort(firstInds);
int[] sortedSecond = Utils.sort(secondInds);
int[] firstCopy = new int[first];
int[] secondCopy = new int[second];
for (int k = 0; k < sortedFirst.length; k++) {
firstCopy[k] = firstInds[sortedFirst[k]];
}
firstInds = firstCopy;
for (int k = 0; k < sortedSecond.length; k++) {
secondCopy[k] = secondInds[sortedSecond[k]];
}
secondInds = secondCopy;
// Unify indices to improve hashing
if (firstInds[0] > secondInds[0]) {
int[] help = secondInds;
secondInds = firstInds;
firstInds = help;
int help2 = second;
second = first;
first = help2;
}
m_Range = new Range(Range.indicesToRangeList(firstInds));
m_Range.setUpper(data.numClasses() - 1);
Range secondRange = new Range(Range.indicesToRangeList(secondInds));
secondRange.setUpper(data.numClasses() - 1);
// Change the class labels and build the classifier
MakeIndicator filter = new MakeIndicator();
filter.setAttributeIndex("" + (data.classIndex() + 1));
filter.setValueIndices(m_Range.getRanges());
filter.setNumeric(false);
filter.setInputFormat(data);
m_FilteredClassifier = new FilteredClassifier();
if (data.numInstances() > 0) {
m_FilteredClassifier.setClassifier(AbstractClassifier.makeCopies(classifier, 1)[0]);
} else {
m_FilteredClassifier.setClassifier(new weka.classifiers.rules.ZeroR());
}
m_FilteredClassifier.setFilter(filter);
// Save reference to hash table at current node
m_classifiers=table;
if (!m_classifiers.containsKey( getString(firstInds) + "|" + getString(secondInds))) {
m_FilteredClassifier.buildClassifier(data);
m_classifiers.put(getString(firstInds) + "|" + getString(secondInds), m_FilteredClassifier);
} else {
m_FilteredClassifier=(FilteredClassifier)m_classifiers.get(getString(firstInds) + "|" +
getString(secondInds));
}
// Create two successors if necessary
m_FirstSuccessor = new DataNearBalancedND();
if (first == 1) {
m_FirstSuccessor.m_Range = m_Range;
} else {
RemoveWithValues rwv = new RemoveWithValues();
rwv.setInvertSelection(true);
rwv.setNominalIndices(m_Range.getRanges());
rwv.setAttributeIndex("" + (data.classIndex() + 1));
rwv.setInputFormat(data);
Instances firstSubset = Filter.useFilter(data, rwv);
m_FirstSuccessor.generateClassifierForNode(firstSubset, m_Range,
rand, classifier, m_classifiers,
instsNumAllClasses);
}
m_SecondSuccessor = new DataNearBalancedND();
if (second == 1) {
m_SecondSuccessor.m_Range = secondRange;
} else {
RemoveWithValues rwv = new RemoveWithValues();
rwv.setInvertSelection(true);
rwv.setNominalIndices(secondRange.getRanges());
rwv.setAttributeIndex("" + (data.classIndex() + 1));
rwv.setInputFormat(data);
Instances secondSubset = Filter.useFilter(data, rwv);
m_SecondSuccessor = new DataNearBalancedND();
m_SecondSuccessor.generateClassifierForNode(secondSubset, secondRange,
rand, classifier, m_classifiers,
instsNumAllClasses);
}
}
/**
* Returns default capabilities of the classifier.
*
* @return the capabilities of this classifier
*/
public Capabilities getCapabilities() {
Capabilities result = super.getCapabilities();
// class
result.disableAllClasses();
result.enable(Capability.NOMINAL_CLASS);
result.enable(Capability.MISSING_CLASS_VALUES);
// instances
result.setMinimumNumberInstances(1);
return result;
}
/**
* Builds tree recursively.
*
* @param data contains the (multi-class) instances
* @throws Exception if the building fails
*/
public void buildClassifier(Instances data) throws Exception {
// can classifier handle the data?
getCapabilities().testWithFail(data);
// remove instances with missing class
data = new Instances(data);
data.deleteWithMissingClass();
Random random = data.getRandomNumberGenerator(m_Seed);
if (!m_hashtablegiven) {
m_classifiers = new Hashtable();
}
// Check which classes are present in the
// data and construct initial list of classes
boolean[] present = new boolean[data.numClasses()];
for (int i = 0; i < data.numInstances(); i++) {
present[(int)data.instance(i).classValue()] = true;
}
StringBuffer list = new StringBuffer();
for (int i = 0; i < present.length; i++) {
if (present[i]) {
if (list.length() > 0) {
list.append(",");
}
list.append(i + 1);
}
}
// Determine the number of instances in each class
double[] instsNum = new double[data.numClasses()];
for (int i = 0; i < data.numInstances(); i++) {
instsNum[(int)data.instance(i).classValue()] += data.instance(i).weight();
}
Range newRange = new Range(list.toString());
newRange.setUpper(data.numClasses() - 1);
generateClassifierForNode(data, newRange, random, m_Classifier, m_classifiers, instsNum);
}
/**
* Predicts the class distribution for a given instance
*
* @param inst the (multi-class) instance to be classified
* @return the class distribution
* @throws Exception if computing fails
*/
public double[] distributionForInstance(Instance inst) throws Exception {
double[] newDist = new double[inst.numClasses()];
if (m_FirstSuccessor == null) {
for (int i = 0; i < inst.numClasses(); i++) {
if (m_Range.isInRange(i)) {
newDist[i] = 1;
}
}
return newDist;
} else {
double[] firstDist = m_FirstSuccessor.distributionForInstance(inst);
double[] secondDist = m_SecondSuccessor.distributionForInstance(inst);
double[] dist = m_FilteredClassifier.distributionForInstance(inst);
for (int i = 0; i < inst.numClasses(); i++) {
if ((firstDist[i] > 0) && (secondDist[i] > 0)) {
System.err.println("Panik!!");
}
if (m_Range.isInRange(i)) {
newDist[i] = dist[1] * firstDist[i];
} else {
newDist[i] = dist[0] * secondDist[i];
}
}
if (!Utils.eq(Utils.sum(newDist), 1)) {
System.err.println(Utils.sum(newDist));
for (int j = 0; j < dist.length; j++) {
System.err.print(dist[j] + " ");
}
System.err.println();
for (int j = 0; j < newDist.length; j++) {
System.err.print(newDist[j] + " ");
}
System.err.println();
System.err.println(inst);
System.err.println(m_FilteredClassifier);
//System.err.println(m_Data);
System.err.println("bad");
}
return newDist;
}
}
/**
* Returns the list of indices as a string.
*
* @param indices the indices to return as string
* @return the indices as string
*/
public String getString(int [] indices) {
StringBuffer string = new StringBuffer();
for (int i = 0; i < indices.length; i++) {
if (i > 0) {
string.append(',');
}
string.append(indices[i]);
}
return string.toString();
}
/**
* @return a description of the classifier suitable for
* displaying in the explorer/experimenter gui
*/
public String globalInfo() {
return
"A meta classifier for handling multi-class datasets with 2-class "
+ "classifiers by building a random data-balanced tree structure.\n\n"
+ "For more info, check\n\n"
+ getTechnicalInformation().toString();
}
/**
* Outputs the classifier as a string.
*
* @return a string representation of the classifier
*/
public String toString() {
if (m_classifiers == null) {
return "DataNearBalancedND: No model built yet.";
}
StringBuffer text = new StringBuffer();
text.append("DataNearBalancedND");
treeToString(text, 0);
return text.toString();
}
/**
* Returns string description of the tree.
*
* @param text the buffer to add the node to
* @param nn the node number
* @return the next node number
*/
private int treeToString(StringBuffer text, int nn) {
nn++;
text.append("\n\nNode number: " + nn + "\n\n");
if (m_FilteredClassifier != null) {
text.append(m_FilteredClassifier);
} else {
text.append("null");
}
if (m_FirstSuccessor != null) {
nn = m_FirstSuccessor.treeToString(text, nn);
nn = m_SecondSuccessor.treeToString(text, nn);
}
return nn;
}
/**
* Returns the revision string.
*
* @return the revision
*/
public String getRevision() {
return RevisionUtils.extract("$Revision: 1.8 $");
}
/**
* Main method for testing this class.
*
* @param argv the options
*/
public static void main(String [] argv) {
runClassifier(new DataNearBalancedND(), argv);
}
}