com.numericalmethod.suanshu.stats.hmm.mixture.distribution

Interface HMMDistribution

All Superinterfaces:

java.io.Serializable

All Known Implementing Classes:

BetaDistribution, BinomialDistribution, ExponentialDistribution, GammaDistribution, LogNormalDistribution, NormalDistribution, PoissonDistribution

public interface HMMDistribution
extends java.io.Serializable

This is the conditional distribution of the observations in each state (possibly differently parameterized) of a mixture hidden Markov model. An EM algorithm, such as the Baum-Welch algorihtm can fit the parameters.

Method Summary

Modifier and Type	Method and Description
ProbabilityDistribution[]	getDistributions() Get the distributions (possibly differently parameterized) for all states.
java.lang.Object[]	getMStepParams(double[] observations, Matrix u, java.lang.Object[] param0) Maximize, for each state, the log-likelihood of the distribution with respect to the observations and current estimators.
java.lang.Object[]	getParams() Get the parameters, for each state, of the distribution.
RandomNumberGenerator[]	getRandomNumberGenerators() Get the random number generators corresponding to the distributions (possibly differently parameterized) for all states.
HMMDistribution	newEMDistribution(java.lang.Object[] lambda) Construct a new distribution from a set of parameters, one set per state.

Method Detail

getParams

java.lang.Object[] getParams()

Get the parameters, for each state, of the distribution.

Returns:

the parameters, for each state, of the distribution

getDistributions

ProbabilityDistribution[] getDistributions()

Get the distributions (possibly differently parameterized) for all states.

Returns:

the distributions

getRandomNumberGenerators

RandomNumberGenerator[] getRandomNumberGenerators()

Get the random number generators corresponding to the distributions (possibly differently parameterized) for all states.

Returns:

the random number generators corresponding to the distributions

getMStepParams

java.lang.Object[] getMStepParams(double[] observations,
                                  Matrix u,
                                  java.lang.Object[] param0)

Maximize, for each state, the log-likelihood of the distribution with respect to the observations and current estimators.

Parameters:

observations - the observations

u - this is a log-transformed version of eq. (4.13) of Zucchini and MacDonald (2009), p. 65

param0 - the current estimators of the distribution parameters

Returns:

the MLE estimators

newEMDistribution

HMMDistribution newEMDistribution(java.lang.Object[] lambda)

Construct a new distribution from a set of parameters, one set per state.

Parameters:

lambda - the parameters of the distribution, one set per state

Returns:

a new HMM distribution