edu.stanford.nlp.maxent.iis
Class Feature

java.lang.Object
  |
  +--edu.stanford.nlp.maxent.iis.Feature

Direct Known Subclasses:

BinaryFeature, ByteFeature, TaggerFeature

public class Feature

extends Object

This class is used as a base class for TaggerFeature for the tagging problem and for BinaryFeature for the general problem with binary features.

Field Summary

protected HashMap	hashValues
int[]	indexedValues This will contain the (x,y) pairs for which the feature is non-zero in case it is sparse.
double	sum
double[]	valuesI These are the non-zero values we want to keep for the points in indexedValues.

Constructor Summary

Feature()

Feature(Experiments e, double[] vals)
This is if we are given an array of double with a value for each training sample in the order of their occurence.

Feature(Experiments e, double[][] vals)

Feature(Experiments e, int[] indexes, double[] vals)

Method Summary

double	ftiled()
double	getVal(int index) Get the value at the index-ed non zero value pair (x,y)
double	getVal(int x, int y) This is rarely used because it is slower and requires initHashVals() to be called beforehand to initiallize the hashValues
int	getX(int index)
int	getY(int index)
void	initHashVals() Creates a hashmap with keys pairs (x,y) and values the value of the function at the pair; required for use of getVal(x,y)
boolean	isEmpty()
int	len()
void	print() Prints out the points where the feature is non-zero and the values at these points.
void	print(PrintStream pf)
void	read(InDataStreamFile inf)
void	save(OutDataStreamFile oF)
void	setSum()
double	sumValues()

Methods inherited from class java.lang.Object

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

Field Detail

indexedValues

public int[] indexedValues

This will contain the (x,y) pairs for which the feature is non-zero in case it is sparse. The pairs (x,y) are coded as x*ySize+y. The values are kept in valuesI. For example, if a feature has only two non-zero values, e.g f(1,2)=3 and f(6,3)=0.74, then indexedValues will have values indexedValues={1*ySize+2,6*ySzie+2} and valuesI will be {3,.74}

valuesI

public double[] valuesI

These are the non-zero values we want to keep for the points in indexedValues.

hashValues

protected HashMap hashValues

sum

public double sum

Constructor Detail

Feature

public Feature()

Feature

public Feature(Experiments e,
               double[] vals)

This is if we are given an array of double with a value for each training sample in the order of their occurence.

Feature

public Feature(Experiments e,
               double[][] vals)

Parameters:

vals - a value for each (x,y) pair

Feature

public Feature(Experiments e,
               int[] indexes,
               double[] vals)

Parameters:

indexes - The pairs (x,y) for which the feature is non-zero. They are coded as x*ySize+y

vals - The values at these points.

Method Detail

print

public void print()

Prints out the points where the feature is non-zero and the values at these points.

print

public void print(PrintStream pf)

getVal

public double getVal(int index)

Get the value at the index-ed non zero value pair (x,y)

setSum

public void setSum()

sumValues

public double sumValues()

save

public void save(OutDataStreamFile oF)

read

public void read(InDataStreamFile inf)

len

public int len()

isEmpty

public boolean isEmpty()

Returns:

true if the feature does not have non-zero values

getX

public int getX(int index)

Returns:

the history x of the index-th (x,y) pair

getY

public int getY(int index)

Returns:

the outcome y of the index-th (x,y) pair

getVal

public double getVal(int x,
                     int y)

This is rarely used because it is slower and requires initHashVals() to be called beforehand to initiallize the hashValues

initHashVals

public void initHashVals()

Creates a hashmap with keys pairs (x,y) and values the value of the function at the pair; required for use of getVal(x,y)

ftiled

public double ftiled()

Returns:

the emptirical expectation of the feature