mulan.classifier.meta
Class HierarchyBuilder

java.lang.Object
  mulan.classifier.meta.HierarchyBuilder

All Implemented Interfaces:

Serializable

public class HierarchyBuilder
extends Object
implements Serializable

Class that builds a hierarchy on flat lables of given mulltilabel data. The hierarchy may be built with three methods.

Version:

0.1

Author:

George Saridis, Grigorios Tsoumakas

See Also:

Serialized Form

Nested Class Summary

static class

HierarchyBuilder.Method The different types of distributing labels to children nodes

Constructor Summary

HierarchyBuilder(int partitions, HierarchyBuilder.Method method)
Constructs a new istance based on given number of partitions and method

Method Summary

MultiLabelInstances	buildHierarchy(MultiLabelInstances mlData) Builds a hierarhical multi-label dataset.
MultiLabelInstances	buildHierarchyAndSaveFiles(MultiLabelInstances mlData, String arffName, String xmlName) Builds the hierarchy and constructs auxiliary files
LabelsMetaData	buildLabelHierarchy(MultiLabelInstances mlData) Builds a hierarhy of labels on top of the labels of a flat multi-label dataset, by recursively partitioning the labels into a specified number of partitions.
static MultiLabelInstances	createHierarchicalDataset(MultiLabelInstances mlData, LabelsMetaData metaData) Creates the hierarchical dataset according to the original multilabel instances object and the constructed label hierarchy

Methods inherited from class java.lang.Object

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

Constructor Detail

HierarchyBuilder

public HierarchyBuilder(int partitions,
                        HierarchyBuilder.Method method)

Constructs a new istance based on given number of partitions and method

Parameters:

partitions - the number of partitions

method - the partitioning method

Method Detail

buildHierarchy

public MultiLabelInstances buildHierarchy(MultiLabelInstances mlData)
                                   throws Exception

Builds a hierarhical multi-label dataset. Firstly a random hierarchy is built on top of the labels of a flat multi-label dataset, by recursively randomly partitioning the labels into a specified number of clusters. Then the values for the new "meta-labels" are properly set, so that the hierarchy is respected.

Parameters:

mlData - the multiLabel data on which the new hierarchy will be built

Returns:

the new multiLabel data

Throws:

Exception

buildLabelHierarchy

public LabelsMetaData buildLabelHierarchy(MultiLabelInstances mlData)
                                   throws Exception

Builds a hierarhy of labels on top of the labels of a flat multi-label dataset, by recursively partitioning the labels into a specified number of partitions.

Parameters:

mlData - the multiLabel data on with the new hierarchy will be built

Returns:

a hierarchy of labels

Throws:

Exception

buildHierarchyAndSaveFiles

public MultiLabelInstances buildHierarchyAndSaveFiles(MultiLabelInstances mlData,
                                                      String arffName,
                                                      String xmlName)
                                               throws Exception

Builds the hierarchy and constructs auxiliary files

Parameters:

mlData - the flat training data

arffName - the name of the hierachical data

xmlName - the filename for the hirearchy

Returns:

the hierarchical data

Throws:

Exception

createHierarchicalDataset

public static MultiLabelInstances createHierarchicalDataset(MultiLabelInstances mlData,
                                                            LabelsMetaData metaData)
                                                     throws InvalidDataFormatException

Creates the hierarchical dataset according to the original multilabel instances object and the constructed label hierarchy

Parameters:

mlData - the original multilabel instances

metaData - the metadata of the constructed label hierarchy

Returns:

the produced dataset

Throws:

InvalidDataFormatException