info.ephyra.questionanalysis.atype.minorthird.hierarchical
Class Experimenter

java.lang.Object
  info.ephyra.questionanalysis.atype.minorthird.hierarchical.Experimenter

public class Experimenter
extends java.lang.Object

A utility class for running experiments with different combinations of the property values required by HierarchicalClassifierTrainer.

Version:

2008-02-10

Author:

Justin Betteridge

Field Summary

private java.lang.String[]	featureTypeCombos
private boolean	isInitialized
private edu.cmu.lti.util.Pair<edu.cmu.lti.javelin.util.Language,edu.cmu.lti.javelin.util.Language>	languagePair
private java.lang.String[]	learningCombos
private static org.apache.log4j.Logger	log
private Properties	properties

Constructor Summary

Experimenter(edu.cmu.lti.util.Pair<edu.cmu.lti.javelin.util.Language,edu.cmu.lti.javelin.util.Language> languagePair)

Method Summary

void	initialize() The input properties that must be defined are: learningCombos : A "|"-separated list of comma-separated lists of learning algorithms.
boolean	isInitialized()
static void	main(java.lang.String[] args) Calls runExperiment() or trainAndSave(), depending on whether the --train command-line argument is specified.
void	runExperiments() Runs the experiments specified in the input properties file.
void	setInitialized(boolean isInitialized)

Methods inherited from class java.lang.Object

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

Field Detail

log

private static org.apache.log4j.Logger log

learningCombos

private java.lang.String[] learningCombos

featureTypeCombos

private java.lang.String[] featureTypeCombos

properties

private Properties properties

languagePair

private edu.cmu.lti.util.Pair<edu.cmu.lti.javelin.util.Language,edu.cmu.lti.javelin.util.Language> languagePair

isInitialized

private boolean isInitialized

Constructor Detail

Experimenter

public Experimenter(edu.cmu.lti.util.Pair<edu.cmu.lti.javelin.util.Language,edu.cmu.lti.javelin.util.Language> languagePair)

Method Detail

initialize

public void initialize()
                throws java.lang.Exception

The input properties that must be defined are:

• learningCombos : A "|"-separated list of comma-separated lists of learning algorithms. The outer, "|"-separated list specifies the different hierarchical classifiers to experiment with. The inner, comma-separated list of algorithms specifies the structure of the hierarchical classifier to use for one experiment. The set of valid algorithm names is:
  • KNN
  • KWAY_MIX
  • MAX_ENT
  • BWINNOW_OVA
  • MPERCEPTRON_OVA
  • NBAYES_OVA
  • VPERCEPTRON_OVA
  • ADABOOST_OVA
  • ADABOOST_CB
  • ADABOOST_MFF
  • ADABOOSTL_OVA
  • ADABOOSTL_CB
  • ADABOOSTL_MFF
  • DTREE_OVA
  • DTREE_CB
  • DTREE_MFF
  • NEGBI_OVA
  • NEGBI_CB
  • NEGBI_MFF
  • SVM_OVA
  • SVM_CB
  • SVM_MFF
• featureTypeCombos : A "|"-separated list of comma-separated lists of feature types.

Throws:

java.lang.Exception

isInitialized

public boolean isInitialized()

Returns:

the isInitialized

setInitialized

public void setInitialized(boolean isInitialized)

Parameters:

isInitialized - the isInitialized to set

runExperiments

public void runExperiments()

Runs the experiments specified in the input properties file. The set of experiments run is the cross-product of the sets of algorithm configurations, training/testing dataset configurations, and feature type combinations.

main

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

Calls runExperiment() or trainAndSave(), depending on whether the --train command-line argument is specified. Also requires two properties files as input arguments: one for specifying the experiments to run and one for configuring EnglishFeatureExtractor

Parameters:

args - input arguments: "[--train] <Experimenter-properties> <qa.properties>"

Throws:

java.lang.Exception