edu.illinois.cs.cogcomp.lbjava.learn

Class RandomWeightVector

java.lang.Object

edu.illinois.cs.cogcomp.lbjava.learn.SparseWeightVector

edu.illinois.cs.cogcomp.lbjava.learn.RandomWeightVector

All Implemented Interfaces:

Serializable, Cloneable

Direct Known Subclasses:

BiasedRandomWeightVector

public class RandomWeightVector
extends SparseWeightVector

This weight vector operates similarly to its parent in the class hierarchy, but it halucinates (and sets) random values for weights corresponding to features it has never been asked about before. Thus, if a dot product is calculated on an empty RandomWeightVector, all the features in the feature vector will now have random weights associated with them in weight vector. If a SparseWeightVector.scaledAdd(int[],double[]) is performed, any features in the feature vector lacking a corresponding weight in the weight vector will have a random one assigned before the addition is performed. This is usually not an issue for most algorithms, since dot products are usually performed before deciding how to add, which means all the weights for that feature vector will already be set when the addition is performed. Thus, it will simply appear to the algorithm that this vector had independent, identically distributed random values for all its dimensions when first created.

The random numbers generated by this class are Gaussian with mean 0 and with a user-configurable standard deviation.

Author:

Nick Rizzolo

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected static double	defaultStddev Default value for stddev.
protected int	instanceNumber Remembers the instance number of this instance.
protected Random	random The random number generator for this instance.
protected double	stddev The random numbers that are generated by this class are Gaussian with mean 0 and standard deviation defined by this variable.

Fields inherited from class edu.illinois.cs.cogcomp.lbjava.learn.SparseWeightVector

defaultCapacity, defaultWeight, weights

Constructor Summary

Constructors

Constructor and Description
RandomWeightVector() Sets a default standard deviation.
RandomWeightVector(double s) Sets the specified standard deviation.

Method Summary

Modifier and Type	Method and Description
void	clear() Empties the weight map and resets the random number generator.
SparseWeightVector	emptyClone() Returns a new, empty weight vector with the same parameter settings as this one.
double	getWeight(int featureIndex, double defaultW) Returns the double precision value for the given feature, or sets a random one and returns it if one did not already exist.
void	read(edu.illinois.cs.cogcomp.core.datastructures.vectors.ExceptionlessInputStream in) Reads the representation of a weight vector with this object's run-time type from the given stream, overwriting the data in this object.
void	write(edu.illinois.cs.cogcomp.core.datastructures.vectors.ExceptionlessOutputStream out) Writes the weight vector's internal representation in binary form.
void	write(PrintStream out) Outputs the contents of this vector into the specified PrintStream.
void	write(PrintStream out, Lexicon lex) Outputs the contents of this vector into the specified PrintStream.

Methods inherited from class edu.illinois.cs.cogcomp.lbjava.learn.SparseWeightVector

clone, dot, dot, getWeight, pairwiseMultiply, readWeightVector, scaledAdd, scaledAdd, scaledAdd, scaledMultiply, scaledMultiply, setWeight, setWeight, size, toString, toString, toStringJustWeights, toStringJustWeights

Methods inherited from class java.lang.Object

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

Field Detail

defaultStddev

protected static final double defaultStddev

Default value for stddev.

See Also:

Constant Field Values

stddev

protected double stddev

The random numbers that are generated by this class are Gaussian with mean 0 and standard deviation defined by this variable.

instanceNumber

protected int instanceNumber

Remembers the instance number of this instance.

random

protected Random random

The random number generator for this instance.

Constructor Detail

RandomWeightVector

public RandomWeightVector()

Sets a default standard deviation.

RandomWeightVector

public RandomWeightVector(double s)

Sets the specified standard deviation.

Parameters:

s - The standard deviation.

Method Detail

getWeight

public double getWeight(int featureIndex,
                        double defaultW)

Returns the double precision value for the given feature, or sets a random one and returns it if one did not already exist.

Overrides:

getWeight in class SparseWeightVector

Parameters:

featureIndex - The feature index

defaultW - Unused.

Returns:

The double precision value for the given feature.

clear

public void clear()

Empties the weight map and resets the random number generator. This means that the same "random" values will be filled in for the weights if the same calls to SparseWeightVector.dot(int[],double[],double) and SparseWeightVector.scaledAdd(int[],double[],double,double) are made in the same order.

Overrides:

clear in class SparseWeightVector

write

public void write(PrintStream out)

Outputs the contents of this vector into the specified PrintStream. The string representation is the same as in the super class, except the "Begin" annotation line also contains the value of stddev in parentheses.

Overrides:

write in class SparseWeightVector

Parameters:

out - The stream to write to.

write

public void write(PrintStream out,
                  Lexicon lex)

Outputs the contents of this vector into the specified PrintStream. The string representation is the same as in the super class, except the "Begin" annotation line also contains the value of stddev in parentheses.

Overrides:

write in class SparseWeightVector

Parameters:

out - The stream to write to.

lex - The feature lexicon.

write

public void write(edu.illinois.cs.cogcomp.core.datastructures.vectors.ExceptionlessOutputStream out)

Writes the weight vector's internal representation in binary form.

Overrides:

write in class SparseWeightVector

Parameters:

out - The output stream.

read

public void read(edu.illinois.cs.cogcomp.core.datastructures.vectors.ExceptionlessInputStream in)

Reads the representation of a weight vector with this object's run-time type from the given stream, overwriting the data in this object.

This method is appropriate for reading weight vectors as written by write(ExceptionlessOutputStream).

Overrides:

read in class SparseWeightVector

Parameters:

in - The input stream.

emptyClone

public SparseWeightVector emptyClone()

Returns a new, empty weight vector with the same parameter settings as this one.

Overrides:

emptyClone in class SparseWeightVector

Returns:

An empty weight vector.