lti::hmmTrainer Class Reference

Hidden Markov Model (HMM) trainer. More...

#include <ltiHmmTrainer.h>

Inheritance diagram for lti::hmmTrainer:

[legend]

Collaboration diagram for lti::hmmTrainer:

[legend]

List of all members.

Classes
class	parameters
	the parameters for the class hmmTrainer More...
Public Member Functions
	hmmTrainer ()
	hmmTrainer (const hmmTrainer &other)
virtual	~hmmTrainer ()
virtual const char *	getTypeName () const
void	reset ()
void	consider (const sequence< dvector > &theSeq)
void	consider (const std::vector< sequence< dvector > > &theSequences)
void	apply (hiddenMarkovModel &model)
dvector	getScoreProgressSequence ()
hmmTrainer &	copy (const hmmTrainer &other)
hmmTrainer &	operator= (const hmmTrainer &other)
virtual functor *	clone () const
const parameters &	getParameters () const

---

Detailed Description

Hidden Markov Model (HMM) trainer.

This class creates a hiddenMarkovModel from a given training set of feature vector sequences.

By default a bakis-topology is chosen (0-1-2 transitions), but this can be changed by redefining minJump and maxJump.

The Viterbi training algorithm, aka segmental K-means, and a mixture density estimator are used to determine the HMM parameters, starting from the user-defined parameters.

See: B.-H. Juang and L. Rabiner, "The segmentation K-means algorithm for estimating %parameters of hidden markov models", IEEE ASSP Magazine, no.9, vol.38, pp.1639-1641,1990.

Note: for computational efficiency probabilities p are not explicitly computed but represented by scores (~ -ln(p)). Note that a high score corresponds to a low probability, i.e. it actually resembles a cost.

---

Constructor & Destructor Documentation

lti::hmmTrainer::hmmTrainer

(

)

default constructor

lti::hmmTrainer::hmmTrainer

(

const hmmTrainer &

other

)

copy constructor

Parameters:

other

the object to be copied

virtual lti::hmmTrainer::~hmmTrainer

(

)

[virtual]

destructor

---

Member Function Documentation

void lti::hmmTrainer::apply

(

hiddenMarkovModel &

model

)

Generate hiddenMarkovModel from the data in the internal data store, that has been filled with the consider() members so far.

virtual functor* lti::hmmTrainer::clone

(

)

const [virtual]

returns a pointer to a clone of this functor.

Implements lti::functor.

void lti::hmmTrainer::consider

(

const std::vector< sequence< dvector > > &

theSequences

)

Consider multiple sequences for computing the hidden markov model.

The sequences are added to an internal data store.

void lti::hmmTrainer::consider

(

const sequence< dvector > &

theSeq

)

Consider this sequence for computing the hidden markov model.

The sequence is added to an internal data store.

hmmTrainer& lti::hmmTrainer::copy

(

const hmmTrainer &

other

)

copy data of "other" functor.

Parameters:

other

the functor to be copied

Returns:

a reference to this functor object

Reimplemented from lti::functor.

const parameters& lti::hmmTrainer::getParameters

(

)

const

returns used parameters

Reimplemented from lti::functor.

dvector lti::hmmTrainer::getScoreProgressSequence

(

)

[inline]

Get the progress of the total score over training data.

The returned vector contains the scores of each iteration as a result of the training (apply()).

virtual const char* lti::hmmTrainer::getTypeName

(

)

const [virtual]

returns the name of this type ("hmmTrainer")

Reimplemented from lti::functor.

hmmTrainer& lti::hmmTrainer::operator=

(

const hmmTrainer &

other

)

alias for copy member

Parameters:

other

the functor to be copied

Returns:

a reference to this functor object

Reimplemented from lti::functor.

void lti::hmmTrainer::reset

(

)

Discard sequences considered so far.

---

The documentation for this class was generated from the following file:

• ltiHmmTrainer.h