dragon.ml.seqmodel.crf
Class AbstractTrainer

java.lang.Object
  |
  +--dragon.ml.seqmodel.crf.AbstractCRF
        |
        +--dragon.ml.seqmodel.crf.AbstractTrainer

All Implemented Interfaces:

Trainer

Direct Known Subclasses:

CollinsBasicTrainer, LBFGSBasicTrainer

public abstract class AbstractTrainer

extends AbstractCRF

implements Trainer

Abstract class for training conditional random field

Copyright: Copyright (c) 2005

Company: IST, Drexel University

Version:

1.0

Author:

Davis Zhou

Field Summary

protected boolean	doScaling
protected int	maxIteration
protected static double	xtol

Fields inherited from class dragon.ml.seqmodel.crf.AbstractCRF

edgeGen, featureGenerator, lambda, model

Constructor Summary

AbstractTrainer(ModelGraph model, FeatureGenerator featureGen)

Method Summary

protected void	genStateVector(DoubleDenseMatrix transMatrix, double[] oldStateVector, double[] newStateVector, boolean transpose)
protected void	genStateVectorLog(DoubleDenseMatrix transMatrix, double[] oldStateVector, double[] newStateVector, boolean transpose)
int	getMaxIteration() Gets the number of maximum iterations
boolean	needScaling() Gets the scaling option.
void	setMaxIteration(int maxIteration) Sets the number of maximum iterations
void	setScalingOption(boolean option) Sets the scaling option

Methods inherited from class dragon.ml.seqmodel.crf.AbstractCRF

computeTransMatrix, computeTransMatrix, getFeatureGenerator, getModelGraph, getModelParameter, readModelParameter, saveModelParameter

Methods inherited from class java.lang.Object

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

Methods inherited from interface dragon.ml.seqmodel.crf.Trainer

getFeatureGenerator, getModelGraph, getModelParameter, saveModelParameter, train

Field Detail

xtol

protected static double xtol

doScaling

protected boolean doScaling

maxIteration

protected int maxIteration

Constructor Detail

AbstractTrainer

public AbstractTrainer(ModelGraph model,
                       FeatureGenerator featureGen)

Method Detail

needScaling

public boolean needScaling()

Description copied from interface: Trainer

Gets the scaling option. The likelihood of the sequence may be too small. Thus it may be necessary to scale the likelihood

Specified by:

needScaling in interface Trainer

Returns:

true if the training needs scaling

setScalingOption

public void setScalingOption(boolean option)

Description copied from interface: Trainer

Sets the scaling option

Specified by:

setScalingOption in interface Trainer

Parameters:

option - the scaling option

getMaxIteration

public int getMaxIteration()

Description copied from interface: Trainer

Gets the number of maximum iterations

Specified by:

getMaxIteration in interface Trainer

Returns:

the number of maximum iterations

setMaxIteration

public void setMaxIteration(int maxIteration)

Description copied from interface: Trainer

Sets the number of maximum iterations

Specified by:

setMaxIteration in interface Trainer

Parameters:

maxIteration - the number of maximum iterations

genStateVector

protected void genStateVector(DoubleDenseMatrix transMatrix,
                              double[] oldStateVector,
                              double[] newStateVector,
                              boolean transpose)

genStateVectorLog

protected void genStateVectorLog(DoubleDenseMatrix transMatrix,
                                 double[] oldStateVector,
                                 double[] newStateVector,
                                 boolean transpose)