cc.mallet.fst.semi_supervised
Class CRFOptimizableByGECriteria

java.lang.Object
  cc.mallet.fst.semi_supervised.CRFOptimizableByGECriteria

All Implemented Interfaces:

Optimizable, Optimizable.ByGradientValue, java.io.Serializable

public class CRFOptimizableByGECriteria
extends java.lang.Object
implements Optimizable.ByGradientValue, java.io.Serializable

GE criteria for training a linear chain CRF.

Author:

Gaurav Chandalia, Gregory Druck

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from interface cc.mallet.optimize.Optimizable

Optimizable.ByBatchGradient, Optimizable.ByCombiningBatchGradient, Optimizable.ByGISUpdate, Optimizable.ByGradient, Optimizable.ByGradientValue, Optimizable.ByHessian, Optimizable.ByValue, Optimizable.ByVotedPerceptron

Field Summary

protected double[]	cachedGradient
protected double	cachedValue
protected CRF	crf
protected InstanceList	data
protected GECriteria	geCriteria
protected cc.mallet.fst.semi_supervised.CRFOptimizableByGECriteria.LatticeCreationExecutor	geLatticeExecutor
protected CRF.Factors	gradient
protected Transducer.Incrementor	incrementor
protected double	priorVariance
protected java.util.concurrent.ThreadPoolExecutor	sumLatticeExecutor

Constructor Summary

CRFOptimizableByGECriteria(GECriteria geCriteria, CRF crf, InstanceList ilist, int numThreads)
Initializes the structures.

Method Summary

void	computeGradient(FeatureVectorSequence input, double[][] gammas, double[][][] xis) Fills gradient from a single instance.
void	computeGradient(java.util.Map<java.lang.Integer,SumLattice> lattices) Resets, computes and fills gradient from all instances.
GECriteria	getGECriteria()
int	getNumParameters()
double	getParameter(int index)
void	getParameters(double[] buffer)
double	getValue()
void	getValueGradient(double[] buffer)
void	initialize(java.util.Map<java.lang.Integer,SumLattice> lattices) Initializes the gradient to zero and re-computes expectations for a new iteration.
void	printGradientAbsNorm()
void	setGaussianPriorVariance(double priorVariance)
void	setParameter(int index, double value)
void	setParameters(double[] buffer)
void	shutdown()

Methods inherited from class java.lang.Object

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

Field Detail

data

protected InstanceList data

crf

protected CRF crf

geCriteria

protected GECriteria geCriteria

gradient

protected CRF.Factors gradient

incrementor

protected Transducer.Incrementor incrementor

cachedValue

protected double cachedValue

cachedGradient

protected double[] cachedGradient

priorVariance

protected double priorVariance

geLatticeExecutor

protected transient cc.mallet.fst.semi_supervised.CRFOptimizableByGECriteria.LatticeCreationExecutor geLatticeExecutor

sumLatticeExecutor

protected transient java.util.concurrent.ThreadPoolExecutor sumLatticeExecutor

Constructor Detail

CRFOptimizableByGECriteria

public CRFOptimizableByGECriteria(GECriteria geCriteria,
                                  CRF crf,
                                  InstanceList ilist,
                                  int numThreads)

Initializes the structures.

Parameters:

geCriteria - GE criteria.

crf - Model.

ilist - Data used for training.

Method Detail

shutdown

public void shutdown()

setGaussianPriorVariance

public void setGaussianPriorVariance(double priorVariance)

getGECriteria

public GECriteria getGECriteria()

initialize

public void initialize(java.util.Map<java.lang.Integer,SumLattice> lattices)

Initializes the gradient to zero and re-computes expectations for a new iteration.

Also creates the executor to compute the gradient (if not done yet).

computeGradient

public void computeGradient(FeatureVectorSequence input,
                            double[][] gammas,
                            double[][][] xis)

Fills gradient from a single instance.

computeGradient

public void computeGradient(java.util.Map<java.lang.Integer,SumLattice> lattices)

Resets, computes and fills gradient from all instances.

Analogous to CRFOptimizableByLabelLikelihood.getExpectationValue.

getValue

public double getValue()

Specified by:

getValue in interface Optimizable.ByGradientValue

getValueGradient

public void getValueGradient(double[] buffer)

Specified by:

getValueGradient in interface Optimizable.ByGradientValue

printGradientAbsNorm

public void printGradientAbsNorm()

getNumParameters

public int getNumParameters()

Specified by:

getNumParameters in interface Optimizable

getParameters

public void getParameters(double[] buffer)

Specified by:

getParameters in interface Optimizable

setParameters

public void setParameters(double[] buffer)

Specified by:

setParameters in interface Optimizable

getParameter

public double getParameter(int index)

Specified by:

getParameter in interface Optimizable

setParameter

public void setParameter(int index,
                         double value)

Specified by:

setParameter in interface Optimizable