be.ac.ulg.montefiore.run.jahmm
Class Hmm<O extends Observation>

java.lang.Object
  be.ac.ulg.montefiore.run.jahmm.Hmm<O>

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable

public class Hmm<O extends Observation>
extends java.lang.Object
implements java.io.Serializable, java.lang.Cloneable

Main Hmm class; it implements an Hidden Markov Model. An HMM is composed of:

• states: each state has a given probability of being initial (pi) and an associated observation probability function (opdf). Each state is associated to an index; the first state is numbered 0, the last n-1 (where n is the number of states in the HMM); this number is given as an argument to the various functions to refer to the matching state.
• transition probabilities: that is, the probability of going from state i to state j (a_i,j).

Important objects extensively used with HMMs are Observations, observation sequences and set of observation sequences. An observation sequence is simply a List of Observations (in the right order, the i-th element of the vector being the i-th element of the sequence). A set of observation sequences is a List of such sequences.

See Also:

Serialized Form

Constructor Summary

	Hmm(double[] pi, double[][] a, java.util.List<? extends Opdf<O>> opdfs) Creates a new HMM.
protected	Hmm(int nbStates) Creates a new HMM.
	Hmm(int nbStates, OpdfFactory<? extends Opdf<O>> opdfFactory) Creates a new HMM.

Method Summary

Hmm<O>	clone()
double	getAij(int i, int j) Returns the probability associated with the transition going from state i to state j (ai,j).
Opdf<O>	getOpdf(int stateNb) Returns the opdf associated with a given state.
double	getPi(int stateNb) Returns the pi value associated with a given state.
double	lnProbability(java.util.List<? extends O> oseq) Returns the neperian logarithm of observation sequence's probability given this HMM.
int[]	mostLikelyStateSequence(java.util.List<? extends O> oseq) Returns an array containing the most likely state sequence matching an observation sequence given this HMM.
int	nbStates() Returns the number of states of this HMM.
double	probability(java.util.List<? extends O> oseq) Returns the probability of an observation sequence given this HMM.
double	probability(java.util.List<? extends O> oseq, int[] sseq) Returns the probability of an observation sequence along a state sequence given this HMM.
void	setAij(int i, int j, double value) Sets the probability associated to the transition going from state i to state j (Ai,j).
void	setOpdf(int stateNb, Opdf<O> opdf) Sets the opdf associated with a given state.
void	setPi(int stateNb, double value) Sets the pi value associated with a given state.
java.lang.String	toString() Gives a description of this HMM.
java.lang.String	toString(java.text.NumberFormat nf) Gives a description of this HMM.

Methods inherited from class java.lang.Object

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

Constructor Detail

Hmm

public Hmm(int nbStates,
           OpdfFactory<? extends Opdf<O>> opdfFactory)

Creates a new HMM. Each state has the same pi value and the transition probabilities are all equal.

Parameters:

nbStates - The (strictly positive) number of states of the HMM.

opdfFactory - A pdf generator that is used to build the pdfs associated to each state.

Hmm

public Hmm(double[] pi,
           double[][] a,
           java.util.List<? extends Opdf<O>> opdfs)

Creates a new HMM. All the HMM parameters are given as arguments.

Parameters:

pi - The initial probability values. pi[i] is the initial probability of state i. This array is copied.

a - The state transition probability array. a[i][j] is the probability of going from state i to state j. This array is copied.

opdfs - The observation distributions. opdfs.get(i) is the observation distribution associated with state i. The distributions are not copied.

Hmm

protected Hmm(int nbStates)

Creates a new HMM. The parameters of the created HMM set to null specified and must be set using the appropriate methods.

Parameters:

nbStates - The (strictly positive) number of states of the HMM.

Method Detail

nbStates

public int nbStates()

Returns the number of states of this HMM.

Returns:

The number of states of this HMM.

getPi

public double getPi(int stateNb)

Returns the pi value associated with a given state.

Parameters:

stateNb - A state number such that 0 ≤ stateNb < nbStates()

Returns:

The pi value associated to stateNb.

setPi

public void setPi(int stateNb,
                  double value)

Sets the pi value associated with a given state.

Parameters:

stateNb - A state number such that 0 ≤ stateNb < nbStates().

value - The pi value to associate to state number stateNb

getOpdf

public Opdf<O> getOpdf(int stateNb)

Returns the opdf associated with a given state.

Parameters:

stateNb - A state number such that 0 ≤ stateNb < nbStates().

Returns:

The opdf associated to state stateNb.

setOpdf

public void setOpdf(int stateNb,
                    Opdf<O> opdf)

Sets the opdf associated with a given state.

Parameters:

stateNb - A state number such that 0 ≤ stateNb < nbStates().

opdf - An observation probability function.

getAij

public double getAij(int i,
                     int j)

Returns the probability associated with the transition going from state i to state j (a_i,j).

Parameters:

i - The first state number such that 0 ≤ i < nbStates().

j - The second state number such that 0 ≤ j < nbStates().

Returns:

The probability associated to the transition going from i to state j.

setAij

public void setAij(int i,
                   int j,
                   double value)

Sets the probability associated to the transition going from state i to state j (A_i,j).

Parameters:

i - The first state number such that 0 ≤ i < nbStates().

j - The second state number such that 0 ≤ j < nbStates().

value - The value of A_i,j.

mostLikelyStateSequence

public int[] mostLikelyStateSequence(java.util.List<? extends O> oseq)

Returns an array containing the most likely state sequence matching an observation sequence given this HMM. This sequence I maximizes the probability of P[I|O,Model] where O is the observation sequence and Model this HMM model.

Parameters:

oseq - A non-empty observation sequence.

Returns:

An array containing the most likely sequence of state numbers. This array can be modified.

probability

public double probability(java.util.List<? extends O> oseq)

Returns the probability of an observation sequence given this HMM.

Parameters:

oseq - A non-empty observation sequence.

Returns:

The probability of this sequence.

lnProbability

public double lnProbability(java.util.List<? extends O> oseq)

Returns the neperian logarithm of observation sequence's probability given this HMM. A scaling procedure is used in order to avoid underflows when computing the probability of long sequences.

Parameters:

oseq - A non-empty observation sequence.

Returns:

The probability of this sequence.

probability

public double probability(java.util.List<? extends O> oseq,
                          int[] sseq)

Returns the probability of an observation sequence along a state sequence given this HMM.

Parameters:

oseq - A non-empty observation sequence.

sseq - An array containing a sequence of state numbers. The length of this array must be equal to the length of oseq

Returns:

The probability P[oseq,sseq|H], where H is this HMM.

toString

public java.lang.String toString(java.text.NumberFormat nf)

Gives a description of this HMM.

Parameters:

nf - A number formatter used to print numbers (e.g. Aij values).

Returns:

A textual description of this HMM.

toString

public java.lang.String toString()

Gives a description of this HMM.

Overrides:

toString in class java.lang.Object

Returns:

A textual description of this HMM.

clone

public Hmm<O> clone()
                                 throws java.lang.CloneNotSupportedException

Overrides:

clone in class java.lang.Object

Throws:

java.lang.CloneNotSupportedException