LBJ2.parse
Class FoldParser

java.lang.Object
  LBJ2.parse.FoldParser

All Implemented Interfaces:

Parser

public class FoldParser
extends java.lang.Object
implements Parser

Useful when performing k-fold cross validation, this parser filters the examples coming from another parser. Conceptually, the examples from the original parser are first split into k "folds" (or partitions) depending on the selected splitting strategy. A particular fold is then selected as the pivot, and this parser can be configured either to return all and only the examples from that fold, or all and only the examples from other folds.

The k folds are referred to by their indexes, which are 0, 1, ..., k - 1. This index is used to select the pivot fold.

See Also:

FoldParser.SplitStrategy

Nested Class Summary

static class

FoldParser.SplitStrategy Immutable type representing the way in which examples are partitioned into folds.

Field Summary

protected int	exampleIndex Keeps track of the index of the next example to be returned.
protected int	examples The total number of examples coming from parser.
protected int	fold Keeps track of the current fold; used only in manual splitting.
protected boolean	fromPivot Whether examples will come from the pivot fold or not.
protected int	K The total number of folds.
protected int	lowerBound A lower bound for an index relating to the pivot fold.
protected Parser	parser The parser whose examples are being filtered.
protected int	pivot The examples from this fold are exclusively selected for or excluded from the set of examples returned by this parser.
protected int[]	shuffled Used only by the random splitting strategy to remember which example indexes are in which folds.
protected int	shuffleIndex An index pointing into shuffled.
protected FoldParser.SplitStrategy	splitStrategy The way in which examples are partitioned into folds.
protected int	upperBound An upper bound for an index relating to the pivot fold.

Constructor Summary

FoldParser(Parser parser, FoldParser.SplitStrategy split, int pivot, boolean f)
Constructor for when you know neither how many examples are in the data nor K, i.e., how many folds are in the data.

FoldParser(Parser parser, int K, FoldParser.SplitStrategy split, int pivot, boolean f)
Constructor for when you don't know how many examples are in the data.

FoldParser(Parser parser, int K, FoldParser.SplitStrategy split, int pivot, boolean f, int e)
Full constructor.

Method Summary

protected boolean	filter(java.lang.Object example) Convenient for determining if the next example should be returned or not.
int	getK() Retrieves the value of K, which may have been computed in the constructor if the splitting strategy is manual.
protected void	increment(java.lang.Object example) Changes state to reflect retrieval of the next example from the parser.
java.lang.Object	next() Retrieves the next example object.
void	reset() Sets this parser back to the beginning of the raw data.
void	setFromPivot(boolean f) Sets the value of fromPivot, which controls whether examples will be taken from the pivot fold or from all other folds.
void	setPivot(int p) Sets the pivot fold, which also causes parser to be reset.

Methods inherited from class java.lang.Object

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

Field Detail

parser

protected Parser parser

The parser whose examples are being filtered.

K

protected int K

The total number of folds.

splitStrategy

protected FoldParser.SplitStrategy splitStrategy

The way in which examples are partitioned into folds.

pivot

protected int pivot

The examples from this fold are exclusively selected for or excluded from the set of examples returned by this parser.

fromPivot

protected boolean fromPivot

Whether examples will come from the pivot fold or not.

examples

protected int examples

The total number of examples coming from parser.

exampleIndex

protected int exampleIndex

Keeps track of the index of the next example to be returned.

fold

protected int fold

Keeps track of the current fold; used only in manual splitting.

lowerBound

protected int lowerBound

A lower bound for an index relating to the pivot fold. The index variable in question may either be exampleIndex or shuffleIndex.

upperBound

protected int upperBound

An upper bound for an index relating to the pivot fold. The index variable in question may either be exampleIndex or shuffleIndex.

shuffled

protected int[] shuffled

Used only by the random splitting strategy to remember which example indexes are in which folds.

shuffleIndex

protected int shuffleIndex

An index pointing into shuffled.

Constructor Detail

FoldParser

public FoldParser(Parser parser,
                  int K,
                  FoldParser.SplitStrategy split,
                  int pivot,
                  boolean f)

Constructor for when you don't know how many examples are in the data. Using a constructor that allows specification of the number of examples in the data only saves computation when the splitting strategy is either sequential or random.

Parameters:

parser - The parser whose examples are being filtered.

K - The total number of folds; this value is ignored if the splitting strategy is manual.

split - The way in which examples are partitioned into folds.

pivot - The index of the pivot fold.

f - Whether to extract examples from the pivot.

FoldParser

public FoldParser(Parser parser,
                  FoldParser.SplitStrategy split,
                  int pivot,
                  boolean f)

Constructor for when you know neither how many examples are in the data nor K, i.e., how many folds are in the data. This constructor can only be used when the splitting strategy is manual. Using a constructor that allows specification of the number of examples in the data only saves computation when the splitting strategy is either sequential or random.

Parameters:

parser - The parser whose examples are being filtered.

split - The way in which examples are partitioned into folds.

pivot - The index of the pivot fold.

f - Whether to extract examples from the pivot.

FoldParser

public FoldParser(Parser parser,
                  int K,
                  FoldParser.SplitStrategy split,
                  int pivot,
                  boolean f,
                  int e)

Full constructor.

Parameters:

parser - The parser whose examples are being filtered.

K - The total number of folds; this value is ignored if the splitting strategy is manual.

split - The way in which examples are partitioned into folds.

pivot - The index of the pivot fold.

f - Whether to extract examples from the pivot.

e - The total number of examples coming from parser, or -1 if unknown.

Method Detail

getK

public int getK()

Retrieves the value of K, which may have been computed in the constructor if the splitting strategy is manual.

setFromPivot

public void setFromPivot(boolean f)

Sets the value of fromPivot, which controls whether examples will be taken from the pivot fold or from all other folds.

Parameters:

f - The new value for fromPivot.

setPivot

public void setPivot(int p)

Sets the pivot fold, which also causes parser to be reset.

Parameters:

p - The index of the new pivot fold.

reset

public void reset()

Sets this parser back to the beginning of the raw data. This means arranging for all relevant state variables to be reset appropriately as well, since the value of pivot may have changed.

Specified by:

reset in interface Parser

See Also:

setPivot(int)

filter

protected boolean filter(java.lang.Object example)

Convenient for determining if the next example should be returned or not.

Parameters:

example - The next example object.

Returns:

true iff the next example should be returned.

increment

protected void increment(java.lang.Object example)

Changes state to reflect retrieval of the next example from the parser.

Parameters:

example - The previous example object.

next

public java.lang.Object next()

Retrieves the next example object.

Specified by:

next in interface Parser

Returns:

The next object parsed from the input data.