cc.mallet.fst

Class Transducer

java.lang.Object

cc.mallet.fst.Transducer

All Implemented Interfaces:

java.io.Serializable

Direct Known Subclasses:

CRF, FeatureTransducer, HMM, PRAuxiliaryModel

public abstract class Transducer
extends java.lang.Object
implements java.io.Serializable

A base class for all sequence models, analogous to classify.Classifier.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static interface	Transducer.Incrementor Methods to be called by inference methods to indicate partial counts of sufficient statistics.
static class	Transducer.State An abstract class used to represent the states of the transducer.
static class	Transducer.TransitionIterator An abstract class to iterate over the states of the transducer.

Field Summary

Fields

Modifier and Type	Field and Description
static double	CERTAIN_WEIGHT
static double	IMPOSSIBLE_WEIGHT

Constructor Summary

Constructors

Constructor and Description
Transducer() Initializes default sum-product and max-product inference engines.
Transducer(Pipe inputPipe, Pipe outputPipe)

Method Summary

Modifier and Type	Method and Description
double	averageTokenAccuracy(InstanceList ilist) Runs inference across all the instances and returns the average token accuracy.
boolean	canIterateAllTransitions() Some transducers are "generative", meaning that you can get a sequence out of them without giving them an input sequence.
SequencePairAlignment	generatePath()
Pipe	getInputPipe()
MaxLatticeFactory	getMaxLatticeFactory()
Pipe	getOutputPipe()
abstract Transducer.State	getState(int index)
SumLatticeFactory	getSumLatticeFactory()
abstract java.util.Iterator	initialStateIterator()
boolean	isGenerative() If true, this is a "generative transducer".
Instance	label(Instance instance) Take input sequence from instance.data and put the output sequence in instance.target.
static double	less_efficient_sumLogProb(double a, double b)
static double	no_longer_needed_sumNegLogProb(double a, double b)
abstract int	numStates()
void	print()
void	setMaxLatticeFactory(MaxLatticeFactory vf)
void	setSumLatticeFactory(SumLatticeFactory fbf)
int	stateIndexOfString(java.lang.String s) Returns the index of the input state name, returns -1 if name not found.
static double	sumLogProb(double a, double b) Returns Math.log(Math.exp(a) + Math.exp(b)).
Instance	transduce(Instance instance) Take input sequence from instance.data and put the output sequence in instance.data.
Sequence	transduce(Sequence input) Converts the given sequence into another sequence according to this transducer.

Methods inherited from class java.lang.Object

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

Field Detail

CERTAIN_WEIGHT

public static final double CERTAIN_WEIGHT

See Also:

Constant Field Values

IMPOSSIBLE_WEIGHT

public static final double IMPOSSIBLE_WEIGHT

See Also:

Constant Field Values

Constructor Detail

Transducer

public Transducer()

Initializes default sum-product and max-product inference engines.

Transducer

public Transducer(Pipe inputPipe,
                  Pipe outputPipe)

Method Detail

getInputPipe

public Pipe getInputPipe()

getOutputPipe

public Pipe getOutputPipe()

setSumLatticeFactory

public void setSumLatticeFactory(SumLatticeFactory fbf)

setMaxLatticeFactory

public void setMaxLatticeFactory(MaxLatticeFactory vf)

getSumLatticeFactory

public SumLatticeFactory getSumLatticeFactory()

getMaxLatticeFactory

public MaxLatticeFactory getMaxLatticeFactory()

label

public Instance label(Instance instance)

Take input sequence from instance.data and put the output sequence in instance.target. Like transduce(Instance), but put best output sequence into instance.target rather than instance.data.

transduce

public Instance transduce(Instance instance)

Take input sequence from instance.data and put the output sequence in instance.data.

transduce

public Sequence transduce(Sequence input)

Converts the given sequence into another sequence according to this transducer. For exmaple, probabilistic transducer may do something like Viterbi here. Subclasses of transducer may specify that they only accept special kinds of sequence.

Parameters:

input - Input sequence

Returns:

Sequence output by this transudcer

numStates

public abstract int numStates()

getState

public abstract Transducer.State getState(int index)

initialStateIterator

public abstract java.util.Iterator initialStateIterator()

canIterateAllTransitions

public boolean canIterateAllTransitions()

Some transducers are "generative", meaning that you can get a sequence out of them without giving them an input sequence. In this case State.transitionIterator() should return all available transitions, but attempts to obtain the input and weight fields may throw an exception.

isGenerative

public boolean isGenerative()

If true, this is a "generative transducer". In this case State.transitionIterator() should return transitions that have valid input and cost fields. True returned here should imply that canIterateAllTransitions() is true.

averageTokenAccuracy

public double averageTokenAccuracy(InstanceList ilist)

Runs inference across all the instances and returns the average token accuracy.

generatePath

public SequencePairAlignment generatePath()

stateIndexOfString

public int stateIndexOfString(java.lang.String s)

Returns the index of the input state name, returns -1 if name not found.

print

public void print()

no_longer_needed_sumNegLogProb

public static double no_longer_needed_sumNegLogProb(double a,
                                                    double b)

sumLogProb

public static double sumLogProb(double a,
                                double b)

Returns Math.log(Math.exp(a) + Math.exp(b)).

a, b represent weights.

less_efficient_sumLogProb

public static double less_efficient_sumLogProb(double a,
                                               double b)