mark.nlp.classifiers
Class HierarchyGenerating

java.lang.Object
  |
  +--mark.nlp.classifiers.HierarchicalClassifier
        |
        +--mark.nlp.classifiers.HierarchyGenerating

All Implemented Interfaces:

Classifier

public class HierarchyGenerating

extends HierarchicalClassifier

A classifier that generates a clustering of the categories and then does a two-level hierarchical training and classifcation. HierarchyGenerating first trains a classifier c_0. Using c_0, HierarchyGenerating generates a confusion matrix for the training set. It then analyzes the confusion matrix. It clusters categories together based on how often they are confused. It next trains a new classifier, h_0, that can place an instance into one of the clusters. For each cluster, it then trains an additional "2nd level" classifier that can distinguish categories within the cluster. At this point, Hierarchy generating has trained c_0, h_0, and a 2nd level classifier for each cluster. It does not use c_0 again. During classification, HierarchyGenerating first sends the instance trhough h_0. Based on h_0's prediction, it then hands off the instance to one of the 2nd level classifiers and returns the prediction of that second-level classifier. It is actually fairly easy to generate a hierarchy more than two levels. For example, to achieve a three level classifier, make your second level classifier another HierarchyGenerating. For that second HierarchyGenerating, make the 2nd level classifier a WekaClassifier. The parameter for this classifier is:

   (
     c0Classifier              // c0's name
     c0ClassifierParameter     // c0's parameter
     numClusters               // the number of clusters
     featureOrdererName0       // h0's orderer
     featureOrdererParameter0  // h0's orderer parameter
     numFeatures0              // h0's number of features
     classifierName0           // h0's name
     classifierParameter0      // h0's parameter
     featureOrdererName1       // 2nd level orderer
     featureOrdererParameter1  // 2nd level orderer parameter
     numFeatures1              // 2nd level number of features
     classifierName1           // 2nd level classifier name
     classifierParameter1      // 2nd level classifier parameter
   )
 

Field Summary

Fields inherited from class mark.nlp.classifiers.HierarchicalClassifier

fC0, fChildren, fHC, fNumCats

Constructor Summary

HierarchyGenerating()

Method Summary

protected mark.nlp.classifiers.Node

genHierarchy(ObjectMap vocabulary, mark.nlp.classifiers.Node confusionClassifier, SimpleResults trainResults, FileCategorySet catSet, SparseBagInstance[] trainSparseBagInstances, java.lang.Object parameters)

Methods inherited from class mark.nlp.classifiers.HierarchicalClassifier

classify, genSubsetNode, instanceFactoryName, test, test, train

Methods inherited from class java.lang.Object

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

Constructor Detail

HierarchyGenerating

public HierarchyGenerating()

Method Detail

genHierarchy

protected mark.nlp.classifiers.Node genHierarchy(ObjectMap vocabulary,
                                                 mark.nlp.classifiers.Node confusionClassifier,
                                                 SimpleResults trainResults,
                                                 FileCategorySet catSet,
                                                 SparseBagInstance[] trainSparseBagInstances,
                                                 java.lang.Object parameters)
                                          throws java.lang.Exception

Specified by:

genHierarchy in class HierarchicalClassifier

java.lang.Exception