edu.cmu.minorthird.classify.algorithms.linear
Class MaxEntLearner

java.lang.Object
  edu.cmu.minorthird.classify.BatchClassifierLearner
      edu.cmu.minorthird.classify.algorithms.linear.MaxEntLearner

All Implemented Interfaces:

ClassifierLearner, java.lang.Cloneable

Direct Known Subclasses:

Recommended.MaxEntLearner

public class MaxEntLearner
extends BatchClassifierLearner

Maximum entropy learner. Based on calling the CRFLearner with sequences of length 1.

Author:

William cohen

Nested Class Summary

static class

MaxEntLearner.MyClassifier

Field Summary

boolean

logSpace

Fields inherited from class edu.cmu.minorthird.classify.BatchClassifierLearner

classifier, dataset

Constructor Summary

MaxEntLearner()

MaxEntLearner(java.lang.String args)
String is list of parameter-value pairs, e.g.

Method Summary

Classifier	batchTrain(Dataset dataset) subclasses should use this method to implement a batch supervised learning algorithm.
boolean	getLogSpace()
ExampleSchema	getSchema() Returns the ExampleSchema - constraints on what the Examples will be.
void	setLogSpace(boolean b)
void	setSchema(ExampleSchema schema) Accept an ExampleSchema - constraints on what the Examples will be.

Methods inherited from class edu.cmu.minorthird.classify.BatchClassifierLearner

addExample, completeTraining, copy, getClassifier, hasNextQuery, nextQuery, reset, setInstancePool

Methods inherited from class java.lang.Object

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

Field Detail

logSpace

public boolean logSpace

Constructor Detail

MaxEntLearner

public MaxEntLearner()

MaxEntLearner

public MaxEntLearner(java.lang.String args)

String is list of parameter-value pairs, e.g. "maxIters 20 mForHessian 5".

Allowed parameters:

• doScaling
• epsForConvergence: Convergence criteria for finding optimum lambda using BFGS
• initValue: initial value for all the lambda arrays
• invSigmaSquare: penalty term for likelihood function is ||lambda||^2*invSigmaSquare/2 set this to zero, if no penalty needed
• maxIters: Maximum number of iterations over the training data during training
• mForHessian: The number of corrections used in the BFGS update.
• trainerType: ...
• scaleScores: if 1, scale scores by assuming each class's hyperplane is log-odds of belonging to that class

For more info, see the docs for the iitb.CRF package.

Method Detail

setLogSpace

public void setLogSpace(boolean b)

getLogSpace

public boolean getLogSpace()

setSchema

public void setSchema(ExampleSchema schema)

Description copied from interface: ClassifierLearner

Accept an ExampleSchema - constraints on what the Examples will be.

getSchema

public ExampleSchema getSchema()

Description copied from interface: ClassifierLearner

Returns the ExampleSchema - constraints on what the Examples will be.

batchTrain

public Classifier batchTrain(Dataset dataset)

Description copied from class: BatchClassifierLearner

subclasses should use this method to implement a batch supervised learning algorithm.

Specified by:

batchTrain in class BatchClassifierLearner