cc.mallet.fst
Class CRF.Factors

java.lang.Object
  cc.mallet.fst.CRF.Factors

All Implemented Interfaces:

java.io.Serializable

Enclosing class:

CRF

public static class CRF.Factors
extends java.lang.Object
implements java.io.Serializable

A simple, transparent container to hold the parameters or sufficient statistics for the CRF.

See Also:

Serialized Form

Nested Class Summary

class	CRF.Factors.Incrementor Instances of this inner class can be passed to various inference methods, which can then gather/increment sufficient statistics counts into the containing Factor instance.
class	CRF.Factors.WeightedIncrementor

Field Summary

double[]	defaultWeights
double[]	finalWeights
double[]	initialWeights
Alphabet	weightAlphabet
SparseVector[]	weights
boolean[]	weightsFrozen

Constructor Summary

CRF.Factors()
Construct a new empty Factors with a new empty weightsAlphabet, 0-length initialWeights and finalWeights, and the other arrays null.

CRF.Factors(CRF.Factors other)
Construct new Factors by mimicking the structure of the other one, but with zero values.

CRF.Factors(CRF.Factors other, boolean cloneAlphabet)
Construct new Factors by copying the other one.

CRF.Factors(CRF crf)
Construct a new Factors with the same structure as the parameters of 'crf', but with values initialized to zero.

Method Summary

void	assertNotNaN()
void	assertNotNaNOrInfinite()
double	gaussianPrior(double variance) Return the log(p(parameters)) according to a zero-mean Gaussian with given variance.
int	getNumFactors()
double	getParameter(int index)
void	getParameters(double[] buffer)
double	getParametersAbsNorm()
double	hyberbolicPrior(double slope, double sharpness) Return the log(p(parameters)) according to a a hyperbolic curve that is a smooth approximation to an L1 prior.
void	plusEquals(CRF.Factors other, double factor)
void	plusEquals(CRF.Factors other, double factor, boolean obeyWeightsFrozen)
void	plusEqualsGaussianPriorGradient(CRF.Factors other, double variance)
void	plusEqualsHyperbolicPriorGradient(CRF.Factors other, double slope, double sharpness)
void	setParameter(int index, double value)
void	setParameters(double[] buff)
boolean	structureMatches(CRF.Factors other)
void	zero()

Methods inherited from class java.lang.Object

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

Field Detail

weightAlphabet

public Alphabet weightAlphabet

weights

public SparseVector[] weights

defaultWeights

public double[] defaultWeights

weightsFrozen

public boolean[] weightsFrozen

initialWeights

public double[] initialWeights

finalWeights

public double[] finalWeights

Constructor Detail

CRF.Factors

public CRF.Factors()

Construct a new empty Factors with a new empty weightsAlphabet, 0-length initialWeights and finalWeights, and the other arrays null.

CRF.Factors

public CRF.Factors(CRF.Factors other)

Construct new Factors by mimicking the structure of the other one, but with zero values. Always simply point to the other's Alphabet; do not clone it.

CRF.Factors

public CRF.Factors(CRF.Factors other,
                   boolean cloneAlphabet)

Construct new Factors by copying the other one.

CRF.Factors

public CRF.Factors(CRF crf)

Construct a new Factors with the same structure as the parameters of 'crf', but with values initialized to zero. This method is typically used to allocate storage for sufficient statistics, expectations, constraints, etc.

Method Detail

getNumFactors

public int getNumFactors()

zero

public void zero()

structureMatches

public boolean structureMatches(CRF.Factors other)

assertNotNaN

public void assertNotNaN()

assertNotNaNOrInfinite

public void assertNotNaNOrInfinite()

plusEquals

public void plusEquals(CRF.Factors other,
                       double factor)

plusEquals

public void plusEquals(CRF.Factors other,
                       double factor,
                       boolean obeyWeightsFrozen)

gaussianPrior

public double gaussianPrior(double variance)

Return the log(p(parameters)) according to a zero-mean Gaussian with given variance.

plusEqualsGaussianPriorGradient

public void plusEqualsGaussianPriorGradient(CRF.Factors other,
                                            double variance)

hyberbolicPrior

public double hyberbolicPrior(double slope,
                              double sharpness)

Return the log(p(parameters)) according to a a hyperbolic curve that is a smooth approximation to an L1 prior.

plusEqualsHyperbolicPriorGradient

public void plusEqualsHyperbolicPriorGradient(CRF.Factors other,
                                              double slope,
                                              double sharpness)

getParametersAbsNorm

public double getParametersAbsNorm()

getParameters

public void getParameters(double[] buffer)

getParameter

public double getParameter(int index)

setParameters

public void setParameters(double[] buff)

setParameter

public void setParameter(int index,
                         double value)