com.exalead.mot.components

Class CRFTrainer

java.lang.Object

com.exalead.gni.GNIObject

com.exalead.mot.components.CRFTrainer

public class CRFTrainer
extends com.exalead.gni.GNIObject

Trains a CRF resource.

Constructor Summary

Constructors

Constructor and Description
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen) CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order) CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order, int mHess) CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order, int mHess, double epsForConvergence) CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order, int mHess, double epsForConvergence, double sigmaSquare) CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order, int mHess, double epsForConvergence, double sigmaSquare, int maxLinesearch) CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int
CRFTrainer(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order, int mHess, double epsForConvergence, double sigmaSquare, int maxLinesearch, int nbThreads) Create a new CRF Trainer.

Method Summary

Modifier and Type	Method and Description
boolean	addSequence(java.lang.String text, int lang, java.lang.String[] labels) Adds a training sequence for string text.
void	dumpTrainBase(java.lang.String dumpFile) Dump the training base to text.
protected void	finalize()
protected void	init(java.lang.String modelDir, MOTPipe pipe, CRFCPGenerator cpGen, int order, int mHess, double epsForConvergence, double sigmaSquare, int maxLinesearch, int nbThreads)
void	train() Launch training on current added sequences.
void	train(int maxIter) Launch training on current added sequences.

Methods inherited from class com.exalead.gni.GNIObject

printInternalCppPointer

Methods inherited from class java.lang.Object

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

Constructor Detail

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen,
                  int order,
                  int mHess,
                  double epsForConvergence,
                  double sigmaSquare,
                  int maxLinesearch,
                  int nbThreads)

Create a new CRF Trainer.

Parameters:

modelDir - the path to the directory which will contain our resource.

pipe - the MOTpipe used for tokenization and context predicate generation.

cpGen - the CRFCPGenerator.

order - the CRF order (only order 1 is supported for the moment).

mHess - number of correction to approximate the inverse hessian matrix.

epsForConvergence - convergence delta (if the difference of log likelihood between to iterations is smaller than this, stop)

sigmaSquare - penalty term for likelihood function.

maxLinesearch - maximum number of trials for the linesearch.

nbThreads - number of threads to use.

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen,
                  int order,
                  int mHess,
                  double epsForConvergence,
                  double sigmaSquare,
                  int maxLinesearch)

CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen,
                  int order,
                  int mHess,
                  double epsForConvergence,
                  double sigmaSquare)

CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen,
                  int order,
                  int mHess,
                  double epsForConvergence)

CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen,
                  int order,
                  int mHess)

CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen,
                  int order)

CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int

CRFTrainer

public CRFTrainer(java.lang.String modelDir,
                  MOTPipe pipe,
                  CRFCPGenerator cpGen)

CRFTrainer(String, MOTPipe, CRFCPGenerator, int, int, double, double, int, int

Method Detail

addSequence

public boolean addSequence(java.lang.String text,
                           int lang,
                           java.lang.String[] labels)

Adds a training sequence for string text.

Parameters:

text - the training text

lang - the language

labels - array of labels for each token in text

Returns:

false if sequence could not be added (ie. not the same number of labels as tokens)

train

public void train()

Launch training on current added sequences.

train

public void train(int maxIter)

Launch training on current added sequences.

Parameters:

maxIter - the maximum number of iterations to perform.

dumpTrainBase

public void dumpTrainBase(java.lang.String dumpFile)

Dump the training base to text.

Parameters:

dumpFile - filename to which we write.

init

protected void init(java.lang.String modelDir,
                    MOTPipe pipe,
                    CRFCPGenerator cpGen,
                    int order,
                    int mHess,
                    double epsForConvergence,
                    double sigmaSquare,
                    int maxLinesearch,
                    int nbThreads)

finalize

protected void finalize()

Overrides:

finalize in class java.lang.Object