ir.classifiers
Class CVLearningCurve

java.lang.Object
  ir.classifiers.CVLearningCurve

public class CVLearningCurve
extends java.lang.Object

Gives learning curves with K-fold cross validation for a classifier.

Field Summary

protected Classifier	classifier The classifier for which K-fold CV learning curve has to be generated
protected boolean	debug Flag for debug display
protected static double[]	DEFAULT_POINTS Default points
protected java.util.Vector<Example>[][]	foldBins foldBins[i][j] stores the examples for class i in fold j.
protected int	numClasses Number of classes in the data
protected int	numFolds Number of folds of cross validation to run
protected double[]	points Points on the X axis (percentage of train data) to plot
protected long	randomSeed Seed for random number generator
protected PointResults[]	testResults Accuracy results for test data, one PointResults for each point on the curve
protected double	testTime Total Testing time
protected int	testTimeNum Total number of examples tested in test time
protected java.util.Vector<Example>[]	totalExamples Stores all the examples for each class
protected int	totalNumTrain Total number of training examples per fold
protected PointResults[]	trainResults Accuracy results for training data, one PointResults for each point on the curve
protected double	trainTime Total Training time

Constructor Summary

CVLearningCurve(Classifier c, java.util.List<Example> examples)
Creates a CVLearning curve object with 10 folds and default points

CVLearningCurve(int nfolds, Classifier c, java.util.List<Example> examples, double[] points, long randomSeed, boolean debug)
Creates a CVLearning curve object

Method Summary

void	binExamples() Set the fold Bins from the total Examples -- this effectively stores the training-test split
Classifier	getClassifier() Return classifier
java.util.Vector<Example>[][]	getFoldBins() Return the fold Bins
java.util.Vector<Example>	getTestCV(int foldnum) Creates the testing set for one fold of a cross-validation on the dataset.
java.util.Vector[]	getTotalExamples() Return all the examples
java.util.Vector<Example>	getTrainCV(int foldnum, double percent) Creates the training set for one fold of a cross-validation on the dataset.
void	run() Run a CV learning curve test and print total training and test time and generate an averge learning curve plot output files suitable for gunuplot
void	setClassifier(Classifier c) Set the classifier
void	setFoldBins(java.util.Vector<Example>[][] bins) Set the fold Bins
void	setTotalExamples(java.util.List<Example> examples) Sets the totalExamples by partitioning examples into categories to get a stratified sample
void	setTotalExamples(java.util.Vector<Example>[] data) Set all the examples
protected int	sizeOfFold(int foldNum) Computes the total number of examples in given fold
void	trainAndTest() Run training and test for each point to be plotted, gathering a result for each fold.
void	trainAndTestFold(java.util.Vector<Example> train, java.util.Vector<Example> test, int fold, PointResults testPointResults, PointResults trainPointResults) Train and test on given example sets for the given fold:

Methods inherited from class java.lang.Object

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

Field Detail

totalExamples

protected java.util.Vector<Example>[] totalExamples

Stores all the examples for each class

foldBins

protected java.util.Vector<Example>[][] foldBins

foldBins[i][j] stores the examples for class i in fold j. This stores the training-test splits for all the folds

classifier

protected Classifier classifier

The classifier for which K-fold CV learning curve has to be generated

randomSeed

protected long randomSeed

Seed for random number generator

numClasses

protected int numClasses

Number of classes in the data

totalNumTrain

protected int totalNumTrain

Total number of training examples per fold

numFolds

protected int numFolds

Number of folds of cross validation to run

points

protected double[] points

Points on the X axis (percentage of train data) to plot

DEFAULT_POINTS

protected static double[] DEFAULT_POINTS

Default points

debug

protected boolean debug

Flag for debug display

trainTime

protected double trainTime

Total Training time

testTime

protected double testTime

Total Testing time

testTimeNum

protected int testTimeNum

Total number of examples tested in test time

testResults

protected PointResults[] testResults

Accuracy results for test data, one PointResults for each point on the curve

trainResults

protected PointResults[] trainResults

Accuracy results for training data, one PointResults for each point on the curve

Constructor Detail

CVLearningCurve

public CVLearningCurve(int nfolds,
                       Classifier c,
                       java.util.List<Example> examples,
                       double[] points,
                       long randomSeed,
                       boolean debug)

Creates a CVLearning curve object

Parameters:

nfolds - Number of folds of CV to perform

c - Classifier on which to perform K-fold CV

examples - List of examples.

points - Points (in percentage of full train set) to plot on learning curve

debug - Debugging flag to set verbose trace printing

CVLearningCurve

public CVLearningCurve(Classifier c,
                       java.util.List<Example> examples)

Creates a CVLearning curve object with 10 folds and default points

Parameters:

c - Classifier on which to perform K-fold CV

examples - List of examples.

Method Detail

getClassifier

public Classifier getClassifier()

Return classifier

setClassifier

public void setClassifier(Classifier c)

Set the classifier

getTotalExamples

public java.util.Vector[] getTotalExamples()

Return all the examples

setTotalExamples

public void setTotalExamples(java.util.Vector<Example>[] data)

Set all the examples

getFoldBins

public java.util.Vector<Example>[][] getFoldBins()

Return the fold Bins

setFoldBins

public void setFoldBins(java.util.Vector<Example>[][] bins)

Set the fold Bins

setTotalExamples

public void setTotalExamples(java.util.List<Example> examples)

Sets the totalExamples by partitioning examples into categories to get a stratified sample

run

public void run()
         throws java.lang.Exception

Run a CV learning curve test and print total training and test time and generate an averge learning curve plot output files suitable for gunuplot

Throws:

java.lang.Exception

trainAndTest

public void trainAndTest()

Run training and test for each point to be plotted, gathering a result for each fold.

trainAndTestFold

public void trainAndTestFold(java.util.Vector<Example> train,
                             java.util.Vector<Example> test,
                             int fold,
                             PointResults testPointResults,
                             PointResults trainPointResults)

Train and test on given example sets for the given fold:

Parameters:

train - The training dataset vector

test - The testing dataset vector

fold - The current fold number

testPointResults - train accuracy PointResults for this point

trainPointResults - test accuracy PointResults for this point

binExamples

public void binExamples()

Set the fold Bins from the total Examples -- this effectively stores the training-test split

getTrainCV

public java.util.Vector<Example> getTrainCV(int foldnum,
                                            double percent)

Creates the training set for one fold of a cross-validation on the dataset.

Parameters:

foldnum - The fold for which training set is to be constructed

percent - Percentage of examples to use for training in this fold

Returns:

The training data

sizeOfFold

protected int sizeOfFold(int foldNum)

Computes the total number of examples in given fold

getTestCV

public java.util.Vector<Example> getTestCV(int foldnum)

Creates the testing set for one fold of a cross-validation on the dataset.

Parameters:

foldnum - The fold which is to be used as testing data

Returns:

The test data