com.sun.labs.minion.classification
Class KFoldSplitter

java.lang.Object
  com.sun.labs.minion.classification.KFoldSplitter

All Implemented Interfaces:

ResultSplitter

public class KFoldSplitter
extends java.lang.Object
implements ResultSplitter

Provides a K-fold splitter. The results are divided up into K equal sized subsets. At each of K iterations, one subset is withheld from the training set to be the validation set.

Nested Class Summary

class

KFoldSplitter.Doc

Field Summary

protected KFoldSplitter.Doc[][]	allDocs All the docs, paired with the array group each comes from
int	currFold The current fold when iterating
int	foldSize The number of documents in each fold.
int	k
protected static java.lang.String	logTag The tag for this module.
protected int	numArrayGroups The number of array groups in the result set
protected ResultSetImpl	parent The full results, as passed in
protected ResultSetImpl	train The set that should be trained on
protected ResultSetImpl	validate The set that should be used for validation

Constructor Summary

KFoldSplitter()
Default constructor.

Method Summary

int	getMinDocs() Gets the minimum number of docs needed for this splitter to be useful.
ResultSetImpl	getTrainSet() Gets the first of the two subset
ResultSetImpl	getValidateSet() Gets the second of the two subsets
void	init(ResultSetImpl parent, IndexConfig iC) Initializes the class.
boolean	nextSplit() Advances to the next split, if there is one.

Methods inherited from class java.lang.Object

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

Field Detail

k

public int k

foldSize

public int foldSize

The number of documents in each fold.

currFold

public int currFold

The current fold when iterating

train

protected ResultSetImpl train

The set that should be trained on

validate

protected ResultSetImpl validate

The set that should be used for validation

parent

protected ResultSetImpl parent

The full results, as passed in

numArrayGroups

protected int numArrayGroups

The number of array groups in the result set

allDocs

protected KFoldSplitter.Doc[][] allDocs

All the docs, paired with the array group each comes from

logTag

protected static java.lang.String logTag

The tag for this module.

Constructor Detail

KFoldSplitter

public KFoldSplitter()

Default constructor. Setup is performed in the init method.

Method Detail

init

public void init(ResultSetImpl parent,
                 IndexConfig iC)

Description copied from interface: ResultSplitter

Initializes the class. Since implementors of ResultSplitter will be created via reflection, the default constructor is used. After instantiation, this init method is called.

Specified by:

init in interface ResultSplitter

Parameters:

parent - the result set to split up

iC - the index config, possibly containing relevent settings for this splitter

getMinDocs

public int getMinDocs()

Gets the minimum number of docs needed for this splitter to be useful.

Specified by:

getMinDocs in interface ResultSplitter

Returns:

the number of folds

getTrainSet

public ResultSetImpl getTrainSet()

Gets the first of the two subset

Specified by:

getTrainSet in interface ResultSplitter

Returns:

a result set that is a subset of the one passed in

getValidateSet

public ResultSetImpl getValidateSet()

Gets the second of the two subsets

Specified by:

getValidateSet in interface ResultSplitter

Returns:

a result set that is a subset of the one passed in

nextSplit

public boolean nextSplit()

Advances to the next split, if there is one. getFirstSet() and getSecondSet() will return the new splits after this method is called

Specified by:

nextSplit in interface ResultSplitter

Returns:

true if there is another split available