com.ibm.able.beans.bayes
Class NaiveBayes

java.lang.Object
  com.ibm.able.beans.bayes.NaiveBayes

All Implemented Interfaces:

java.io.Serializable

public class NaiveBayes

extends java.lang.Object

implements java.io.Serializable

See Also:

Serialized Form

Field Summary

protected static long

serialVersionUID

Constructor Summary

NaiveBayes()

NaiveBayes(int ncls, int nftr, int[] nval, double m, double[] cpriors, double[][][] ppriors)
construct NaiveBayes with the explicitly specified parameters ncls - number of class labels nftr - number of features nval - number of values per each feature (assuming nominal - discrete finite-valued - features) cpriors - prior probability distribution over class labels m - equivalent sample size ppriors - prior estimates of the probabilities P(f|C) (used for Bayesian parameter estimation with equivalent sample size method)

Method Summary

void	buildHypothesis(int[][] data, int[] labels, int ninst, int ncls, int nftr, int[] nval) Build a hypothesis using explicit parameters This function learns a naive Bayes model given a set of labeled instances.
double	classify(int[][] data, int[] labels, int ninst, boolean[] selectedFeatures) Classify a record Input: data - training data in a table format where each row represents an instance, and each column represents an attribute (feature).
int	classifyExample(int[] instance, boolean[] selectedFeatures) This function selects the maximum-likelihood class label given a data instance Input: instance - feature vector (discrete finite feature values represented by integers) selectedFeatures - boolean array specifying selected features (by default, null -all features included)
double[]	findClassProbability(int[] instance, boolean[] selectedFeatures) This function returns the posterior probability distribution over class labels, given a data instance using Bayes rule as follows: find P(class|instance)=P(instance|class)P(class)/sum_i P(instnace|class_i) Input: instance - feature vector (discrete finite feature values represented by integers) selectedFeatures - boolean array specifying selected features (by default, null -all features included)
double	getAccuracy()
double	getAvgLikelihood()
double	getAvgLogLikelihood()
double[]	getClassPriors()
double[]	getClassProb()
int[][]	getConfusionMatrix()
double[][][]	getCPT()
double[]	getEqSampleSize()
int	getNClasses()
int	getNFeatures()
int[]	getNFValues()
void	initializeNB(int ncls, int nftr, int[] nval, double eqss, double[] cpriors, double[][][] ppriors) internal function that implements the class construction with explicit list of parameters ncls - number of class labels nftr - number of features nval - number of values per each feature (assuming nominal - discrete finite-valued - features) cpriors - prior probability distribution over class labels eqss - equivalent sample sizes for each class (by deafult, each class was seen at least once) ppriors - prior estimates of the probabilities P(f|C) (used for Bayesian parameter estimation with equivalent sample size method)
double	likelihood(int[] instance, int classlabel) Compute the likelihood of an instance given a class label
void	setClassPriors(double[] cpriors)
void	setCPT(double[][][] cptpriors)
void	setNClasses(int ncls)
void	setNFeatures(int nftr, int[] nfv)

Methods inherited from class java.lang.Object

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

Field Detail

serialVersionUID

protected static final long serialVersionUID

See Also:

Constant Field Values

Constructor Detail

NaiveBayes

public NaiveBayes()

NaiveBayes

public NaiveBayes(int ncls,
                  int nftr,
                  int[] nval,
                  double m,
                  double[] cpriors,
                  double[][][] ppriors)
           throws AbleException

construct NaiveBayes with the explicitly specified parameters ncls - number of class labels nftr - number of features nval - number of values per each feature (assuming nominal - discrete finite-valued - features) cpriors - prior probability distribution over class labels m - equivalent sample size ppriors - prior estimates of the probabilities P(f|C) (used for Bayesian parameter estimation with equivalent sample size method)

Method Detail

initializeNB

public void initializeNB(int ncls,
                         int nftr,
                         int[] nval,
                         double eqss,
                         double[] cpriors,
                         double[][][] ppriors)
                  throws AbleException

internal function that implements the class construction with explicit list of parameters ncls - number of class labels nftr - number of features nval - number of values per each feature (assuming nominal - discrete finite-valued - features) cpriors - prior probability distribution over class labels eqss - equivalent sample sizes for each class (by deafult, each class was seen at least once) ppriors - prior estimates of the probabilities P(f|C) (used for Bayesian parameter estimation with equivalent sample size method)

Throws:

AbleException

getNClasses

public int getNClasses()

getNFeatures

public int getNFeatures()

getNFValues

public int[] getNFValues()

getEqSampleSize

public double[] getEqSampleSize()

getCPT

public double[][][] getCPT()

getClassPriors

public double[] getClassPriors()

getAvgLikelihood

public double getAvgLikelihood()

getAvgLogLikelihood

public double getAvgLogLikelihood()

getAccuracy

public double getAccuracy()

getClassProb

public double[] getClassProb()

getConfusionMatrix

public int[][] getConfusionMatrix()

setNClasses

public void setNClasses(int ncls)
                 throws java.lang.IllegalArgumentException

Throws:

java.lang.IllegalArgumentException

setNFeatures

public void setNFeatures(int nftr,
                         int[] nfv)
                  throws java.lang.IllegalArgumentException

Throws:

java.lang.IllegalArgumentException

setCPT

public void setCPT(double[][][] cptpriors)
            throws AbleException

Throws:

AbleException

setClassPriors

public void setClassPriors(double[] cpriors)

buildHypothesis

public void buildHypothesis(int[][] data,
                            int[] labels,
                            int ninst,
                            int ncls,
                            int nftr,
                            int[] nval)

Build a hypothesis using explicit parameters This function learns a naive Bayes model given a set of labeled instances. Given class value C=c, for each attribute (feature) f, it computes an estimate of the parameter P(f|C=c) using the m-estimate of probability, or so-called equivalent sample size method (see Mitchel, Machine Learning): P(f=a|C=c)= (n + mp)/(N+m), where N is the total number of instances for which C=c, n is the number of these instances for which also f=a, p is a prior estimate of the probability P(f=a|C=c), and m is a constant called the equivalent sample size. The m-estimate can be interpreted as an empirical probability estimate (i.e. frequency) assuming m additional "prior" instances distributed according to p, combined with the "current" N instances. When there is no prior information about p, we assume uniform distribution over all possible values for each feature, i.e. p=1/nFValues. Input: data - training data in a table format where each row represents an instance, and each column represents an attribute (feature). labels - the class label. ninst - the number of training instances ncls - number of class labels nftr - number of features nval - number of values per each feature (assuming nominal - discrete finite-valued - features) (We always assume that the class labels and the feature number indexes start from 0).

classify

public double classify(int[][] data,
                       int[] labels,
                       int ninst,
                       boolean[] selectedFeatures)

Classify a record Input: data - training data in a table format where each row represents an instance, and each column represents an attribute (feature). labels - the class label. ninst - the number of training instances selectedFeatures - boolean array specifying selected features (by default, null -all features included) We always assume that the class labels and the feature number indexes start from 0.

classifyExample

public int classifyExample(int[] instance,
                           boolean[] selectedFeatures)

This function selects the maximum-likelihood class label given a data instance Input: instance - feature vector (discrete finite feature values represented by integers) selectedFeatures - boolean array specifying selected features (by default, null -all features included)

findClassProbability

public double[] findClassProbability(int[] instance,
                                     boolean[] selectedFeatures)

This function returns the posterior probability distribution over class labels, given a data instance using Bayes rule as follows: find P(class|instance)=P(instance|class)P(class)/sum_i P(instnace|class_i) Input: instance - feature vector (discrete finite feature values represented by integers) selectedFeatures - boolean array specifying selected features (by default, null -all features included)

likelihood

public double likelihood(int[] instance,
                         int classlabel)
                  throws java.lang.IllegalArgumentException

Compute the likelihood of an instance given a class label

Throws:

java.lang.IllegalArgumentException