weka.classifiers.meta
Class HND

java.lang.Object
  weka.classifiers.Classifier
      weka.classifiers.meta.HND

All Implemented Interfaces:

java.lang.Cloneable, NestedDichotomy, OptionHandler, Randomizable, java.io.Serializable

public class HND

extends Classifier

implements NestedDichotomy, OptionHandler

Class to create levelwise NDs with respect to a given hierarchy of classes. Valid options are:

-P classname
Specify the full class name of a ClassHierarchyParser (default: weka.core.converters.ClassTreeArffFileParser).

-H classHierarchy
A String specifying the class-hierarchy in a valid manner with respect to the specified ClassHierarchyParser (required).

-L property-file
A String specifying the filename providing levelwise options for the respective ND and for selection of attributes (optional).
A key for ND options is supposed to be of form "leveln" for level n. Levelwise properties will overwrite the general options for ND or select attributes for the respective level only.
A key for levelwise selection of attributes is supposed to be of form "removen" for level n. The property for this key is supposed to be a string providing the options for the Filter Remove.

Options after "--" are passed to the inherent NDs.

Author:

Arthur Zimek

See Also:

Serialized Form

Field Summary

private java.util.Map	childrenHNDs The HNDs covering the instances at deeper levels, if there are such.
private ClassHierarchyParser	classHierarchyParser The default parser is a ClassTreeArffFileParser, suitable for a hierarchy encoded within a single String.
private ClassHierarchy	hierarchy The hierarchy of classes for the current and deeper levels.
private int	LEVEL The level of this HND in the hierarchy (root is 0).
static java.lang.String	LEVEL_PROPERTY Key-prefix for the property level.
private ND	nd The ND to classify the instances at the currentlevel
private java.lang.String[]	ndOptions The options to be set for each ND.
private java.util.Properties	properties The properties for levelwise settings of the respective ND.
private java.lang.String	propertyFilename The filename of the property-file as set via setOptions.
static java.lang.String	REMOVE_FILTER_PROPERTY Key-prefix for the property filter, for levelwise selection of attributes.
static java.util.regex.Pattern	WHITESPACE The pattern for separation of entries in the option-line of a property-file.

Fields inherited from class weka.classifiers.Classifier

m_Debug

Constructor Summary

	HND() Default constructor creates a HND without hierarchy for level 0.
private	HND(ClassHierarchy hierarchy, java.util.Properties properties, int level) The constructor to be used by the buildLevelwiseClassifier(Instances) method provides a hierarchy.

Method Summary

private void	appendDescription(java.lang.StringBuffer description) Appends a description of all children HNDs of this HND to the given description.
void	buildClassifier(Instances data) Builds the classifier for the given training data.
private void	buildLevelwiseClassifier(Instances data) Build levelwise NDs with respect to the specified hierarchy of classes.
double[]	distributionForInstance(Instance instance) Returns a class probability estimation for the given instance.
private void	distributionForInstance(Instance instance, java.util.Map classDistribution, double factor) Computes levelwise class probability estimation for the given instance and fills in the respective values to the classDistribution, using the factor (accumulated class probability for the respective superclass).
java.lang.String[]	getOptions() Returns the current settings of the Classifier.
int	getSeed() Returns the current seed of the ND.
void	initND(java.lang.String[] ndOptions) Sets the given options to the ND.
java.util.Enumeration	listOptions() Returns a list of the available options.
static void	main(java.lang.String[] args) Main method for testing this class.
void	setOptions(java.lang.String[] options) Parses a given list of options.
void	setSeed(int seed) Sets the given seed to the ND and recursively to NDs of children HNDs.
java.lang.String	toString() Returns a description of this classifier.

Methods inherited from class weka.classifiers.Classifier

classifyInstance, debugTipText, forName, getDebug, makeCopies, setDebug

Methods inherited from class java.lang.Object

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

Field Detail

LEVEL_PROPERTY

public static final java.lang.String LEVEL_PROPERTY

Key-prefix for the property level.

See Also:

Constant Field Values

REMOVE_FILTER_PROPERTY

public static final java.lang.String REMOVE_FILTER_PROPERTY

Key-prefix for the property filter, for levelwise selection of attributes.

See Also:

Constant Field Values

WHITESPACE

public static final java.util.regex.Pattern WHITESPACE

The pattern for separation of entries in the option-line of a property-file. All single whitespace characters are valid separators.

nd

private ND nd

The ND to classify the instances at the currentlevel

properties

private java.util.Properties properties

The properties for levelwise settings of the respective ND.

propertyFilename

private java.lang.String propertyFilename

The filename of the property-file as set via setOptions.

LEVEL

private final int LEVEL

The level of this HND in the hierarchy (root is 0).

ndOptions

private java.lang.String[] ndOptions

The options to be set for each ND.

childrenHNDs

private java.util.Map childrenHNDs

The HNDs covering the instances at deeper levels, if there are such. Remains null for a HND covering the base-classes (leaves of the hierarchy). Is set during the method buildLevelwiseCLassifier. A superclasses-String is key for the according HND.

hierarchy

private ClassHierarchy hierarchy

The hierarchy of classes for the current and deeper levels.

classHierarchyParser

private ClassHierarchyParser classHierarchyParser

The default parser is a ClassTreeArffFileParser, suitable for a hierarchy encoded within a single String.

Constructor Detail

HND

public HND()

Default constructor creates a HND without hierarchy for level 0. The hierarchy is either to be set via setOptions(String[] or remains null, in which case any data will be rejected as invalid.

HND

private HND(ClassHierarchy hierarchy,
            java.util.Properties properties,
            int level)

The constructor to be used by the buildLevelwiseClassifier(Instances) method provides a hierarchy.

Parameters:

hierarchy - the specified hierarchy for the respective level

properties - the levelwise properties to be set

level - the current level of the hierarchy

Method Detail

initND

public void initND(java.lang.String[] ndOptions)
            throws java.lang.Exception

Sets the given options to the ND.

Parameters:

ndOptions - the options to be set to the ND. The parameter ndOptions is copied, not altered, since it is to be passed through all levels.

Throws:

java.lang.Exception - in case of invalid options for ND

buildLevelwiseClassifier

private void buildLevelwiseClassifier(Instances data)
                               throws java.lang.Exception

Build levelwise NDs with respect to the specified hierarchy of classes.

Parameters:

data - training instances

Throws:

java.lang.Exception

buildClassifier

public void buildClassifier(Instances data)
                     throws java.lang.Exception

Builds the classifier for the given training data.

Specified by:

buildClassifier in class Classifier

Parameters:

data - set of instances serving as training data

Throws:

java.lang.Exception - if the classifier has not been generated successfully

See Also:

Classifier.buildClassifier(Instances)

distributionForInstance

public double[] distributionForInstance(Instance instance)
                                 throws java.lang.Exception

Returns a class probability estimation for the given instance.

Overrides:

distributionForInstance in class Classifier

Parameters:

instance - the instance to be classified

Returns:

an array containing the estimated membership probabilities of the test instance in each class or the numeric prediction

Throws:

java.lang.Exception - if distribution could not be computed successfully

See Also:

Classifier.distributionForInstance(weka.core.Instance)

distributionForInstance

private void distributionForInstance(Instance instance,
                                     java.util.Map classDistribution,
                                     double factor)
                              throws java.lang.Exception

Computes levelwise class probability estimation for the given instance and fills in the respective values to the classDistribution, using the factor (accumulated class probability for the respective superclass).

Parameters:

instance - the instance to be classified

classDistribution - the class distribution covering all classes

factor - the class probability for the respective superclass

Throws:

java.lang.Exception

getOptions

public java.lang.String[] getOptions()

Returns the current settings of the Classifier.

Specified by:

getOptions in interface OptionHandler

Overrides:

getOptions in class Classifier

Returns:

an array of strings suitable for passing to setOptions

See Also:

OptionHandler.getOptions()

listOptions

public java.util.Enumeration listOptions()

Returns a list of the available options.

Specified by:

listOptions in interface OptionHandler

Overrides:

listOptions in class Classifier

Returns:

an enumeration of all the available options.

See Also:

OptionHandler.listOptions()

setOptions

public void setOptions(java.lang.String[] options)
                throws java.lang.Exception

Parses a given list of options. Valid options are:

-P classname
Specify the full class name of a ClassHierarchyParser (default: weka.core.converters.ClassTreeArffFileParser).

-H classHierarchy
A String specifying the class-hierarchy in a valid manner with respect to the specified ClassHierarchyParser (required).

-L property-file
A String specifying the filename providing levelwise options for the respective ND and for selection of attributes (optional).
A key for ND options is supposed to be of form "leveln" for level n. Levelwise properties will overwrite the general options for ND or select attributes for the respective level only.
A key for levelwise selection of attributes is supposed to be of form "removen" for level n. The property for this key is supposed to be a string providing the options for the Filter Remove.

Options after -- are passed to the inherent NDs.

Specified by:

setOptions in interface OptionHandler

Overrides:

setOptions in class Classifier

Parameters:

options - the list of options as an array of strings

Throws:

java.lang.Exception - if a required option is not supported

See Also:

OptionHandler.setOptions(java.lang.String[])

setSeed

public void setSeed(int seed)

Sets the given seed to the ND and recursively to NDs of children HNDs.

Specified by:

setSeed in interface Randomizable

Parameters:

seed - the seed

See Also:

Randomizable.setSeed(int)

getSeed

public int getSeed()

Returns the current seed of the ND.

Specified by:

getSeed in interface Randomizable

Returns:

the seed for the random number generation

See Also:

Randomizable.getSeed()

toString

public java.lang.String toString()

Returns a description of this classifier.

Returns:

String a description of this classifier

See Also:

Object.toString()

appendDescription

private void appendDescription(java.lang.StringBuffer description)

Appends a description of all children HNDs of this HND to the given description.

Parameters:

description - the description to which further descriptions are to be appended

main

public static void main(java.lang.String[] args)

Main method for testing this class.

Parameters:

args - the options