edu.mit.nlp.segmenter.mcmc
Class CuCoSeg.PriorOptimizer

java.lang.Object
  extended by edu.mit.util.weka.LBFGSWrapper
      extended by edu.mit.nlp.segmenter.mcmc.CuCoSeg.PriorOptimizer
Enclosing class:
CuCoSeg

protected class CuCoSeg.PriorOptimizer
extends LBFGSWrapper

An LBFGS optimizer to search the parameter space


Field Summary
 
Fields inherited from class edu.mit.util.weka.LBFGSWrapper
m_debug, m_eps, m_estimate, m_max_its, m_num_corrections, m_num_parameters, m_value, xtol
 
Constructor Summary
CuCoSeg.PriorOptimizer()
           
 
Method Summary
 double computeGradientForSegment(int[] counts, double sum, double prior, FastDigamma digamma)
           
 double[] evaluateGradient(double[] params)
          assumes params are in log form
 double objectiveFunction(double[] params)
          assumes params are in log form
 void scanParams(double[] params)
           
 
Methods inherited from class edu.mit.util.weka.LBFGSWrapper
findArgmin, getMinFunction, getVarbValues, setDebug, setEstimate, setMaxIteration
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

CuCoSeg.PriorOptimizer

public CuCoSeg.PriorOptimizer()
Method Detail

scanParams

public void scanParams(double[] params)

objectiveFunction

public double objectiveFunction(double[] params)
assumes params are in log form

Specified by:
objectiveFunction in class LBFGSWrapper

evaluateGradient

public double[] evaluateGradient(double[] params)
assumes params are in log form

Specified by:
evaluateGradient in class LBFGSWrapper

computeGradientForSegment

public double computeGradientForSegment(int[] counts,
                                        double sum,
                                        double prior,
                                        FastDigamma digamma)


Copyright © 2008 MIT. All Rights Reserved.