marytts.machinelearning
Class GmmDiscretizer

java.lang.Object
  marytts.machinelearning.GmmDiscretizer

All Implemented Interfaces:

Discretizer

public class GmmDiscretizer
extends java.lang.Object
implements Discretizer

This discretizes values according to a gaussian mixture model (gmm). The result of discretization is the mean of the class that contributed most probability to a point.

Author:

benjaminroth

Constructor Summary

GmmDiscretizer(GMM model, boolean extraZeroClass)
This constructs a Discretizer using the specified mixture model.

Method Summary

int	discretize(int value) This discretizes a value by returning the mean of that gaussian component that has maximum probability for it.
int[]	getPossibleValues() Returns all poosible discretizations values can be mapped to.
static GmmDiscretizer	trainDiscretizer(java.util.List<java.lang.Integer> values, int nrClasses, boolean extraZero) This trains a gaussian mixture model having the specified number of components.

Methods inherited from class java.lang.Object

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

Constructor Detail

GmmDiscretizer

public GmmDiscretizer(GMM model,
                      boolean extraZeroClass)

This constructs a Discretizer using the specified mixture model.

Parameters:

model - GMM to be used

extraZeroClass - specifies if zeros should be treated independently

Method Detail

trainDiscretizer

public static GmmDiscretizer trainDiscretizer(java.util.List<java.lang.Integer> values,
                                              int nrClasses,
                                              boolean extraZero)

This trains a gaussian mixture model having the specified number of components.

Parameters:

values - the data the model is trained with

nrClasses - number of components the mixture will have

extraZero - specifies if zeroes are to be treated seperately from mixture model training and application.

Returns:

a discretizer that discretizes according to the trained model

discretize

public int discretize(int value)

This discretizes a value by returning the mean of that gaussian component that has maximum probability for it.

Specified by:

discretize in interface Discretizer

Parameters:

value - the value to be discretized

Returns:

the discretization the value is mapped to

getPossibleValues

public int[] getPossibleValues()

Returns all poosible discretizations values can be mapped to.

Specified by:

getPossibleValues in interface Discretizer

Returns:

all poosible discretizations values can be mapped to.