jbnc
Class CrossVal

java.lang.Object
  jbnc.Classifier
      jbnc.CrossVal

public class CrossVal

extends Classifier

Cross validation classifier tester. Assumes that cross validation files are already generated an attempts to read them using given file stem. The file name format is: "stem-repetition-fold". For instance for stem "monk" the file names could be "monk-0-0.*", "monk-0-1.*", etc.

 USAGE: CrossVal {options}

 where:
  -a {algor_name} Algorithm choices: "naive", "TAN", "FAN", "STAN", "STAND".
  -c {class_name} Name of the class variable. The default value is 'class'.
  -d              Print debugging information.
  -f {file_stem}  Load test data in C4.5 format (.names + .test),
                   file_stem-?-?.names - file with specification of attributes,
                   file_stem-?-?.data  - file with train cases.
                   file_stem-?-?.test  - file with test cases.
  -n {file_name}  Save constructed naive-Bayes networks to file_name-?-?.bif.
                  File is saved in BIF 0.15 format.
  -q {q_measure}  Select quality mesure. This relevant for the inducers that
                  perform selection of network candidates, like FAN or STAN.
                  Diffrent measure have drasticaly diffrent computational
                  complexity.
                  Quality measures choices:
                   LC     - local criterion: log p(c_l|D) [local]
                   SB     - standard Bayesian measure with penalty for size
                            [global].
                   HGS    - Heckerman-Geiger-Chickering measure [global].
                   LOO    - leave-one-out cross validation [local].
                   CV10   - ten fold cross validation.[local].
                   CV1010 - ten fold cross validation averaged ten times
                            [local].
  -s              Number of smoothing priors to test. Has to be an integer
                  greater or equal zero.
  -t              Print execution time.

 EXAMPLE: CrossVal -a TAN -tf monk1
 EXAMPLE: CrossVal -dta FAN -q LOO -f monk1

Since:

June 1, 1999

Author:

Jarek Sacha

Field Summary

Fields inherited from class jbnc.Classifier

algorithmName, className, classNameDefault, databaseURL, debugMode, fileNameStem, measureName, namesFileName, nbAlphas, netFileName, tableName, testFileName, trainFileName, usageMsg, useTimer

Constructor Summary

CrossVal()

Method Summary

protected DatasetInt	loadDataset(java.lang.String namesFile, java.lang.String dataFile) Dataset loading wrapper.
static void	main(java.lang.String[] args) Main function.
protected void	run() Create and test a classifier.
protected void	run(java.lang.String[] args)
protected BNCTester.Result	test(BayesianNetworks.BayesNet net, DatasetInt testSet, boolean report) Bayesian network classifier testing wrapper.
protected BayesianNetworks.BayesNet	train(BayesianInducer inducer, FrequencyCalc fc, boolean usePriors, double alphaK) Bayesian network classifier training wrapper.

Methods inherited from class jbnc.Classifier

processCommandLine, setUsageMsg

Methods inherited from class java.lang.Object

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

Constructor Detail

CrossVal

public CrossVal()

Method Detail

run

protected void run(java.lang.String[] args)

Overrides:

run in class Classifier

main

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

Main function.

Parameters:

args - The command line arguments

train

protected BayesianNetworks.BayesNet train(BayesianInducer inducer,
                                          FrequencyCalc fc,
                                          boolean usePriors,
                                          double alphaK)
                                   throws java.lang.Exception

Bayesian network classifier training wrapper.

Parameters:

inducer - algorithm used for classifier learning.

fc - Description of Parameter

usePriors - Description of Parameter

alphaK - Description of Parameter

Returns:

constructed Bayesian classifier network.

Throws:

java.lang.Exception - Description of Exception

test

protected BNCTester.Result test(BayesianNetworks.BayesNet net,
                                DatasetInt testSet,
                                boolean report)
                         throws java.lang.Exception

Bayesian network classifier testing wrapper.

Parameters:

net - Bayesian network classifier.

testSet - test data.

report - wether to report results on the screen.

Returns:

object containg results of testing.

Throws:

java.lang.Exception - Description of Exception

run

protected void run()

Description copied from class: Classifier

Create and test a classifier.

Overrides:

run in class Classifier

loadDataset

protected DatasetInt loadDataset(java.lang.String namesFile,
                                 java.lang.String dataFile)
                          throws java.lang.Exception

Dataset loading wrapper.

Overrides:

loadDataset in class Classifier

Parameters:

namesFile - Description of Parameter

dataFile - Description of Parameter

Returns:

Description of the Returned Value

Throws:

java.lang.Exception - Description of Exception