Tanl::POS::Viterbi Class Reference

This is the good old Viterbi algorithm for HMM decoding. More...

#include <Viterbi.h>

List of all members.

Classes
struct	Node
Public Member Functions
	Viterbi (HMM &hmm)
std::vector< State >	decode (double logtheta, State &startState, HMM::Observations &observations)
	logtheta is used for beam pruning.

---

Detailed Description

This is the good old Viterbi algorithm for HMM decoding.

Viterbi doesn't deal with the structure of an HMM. It only needs a start state and a black box that calculates the following states with transition and emission probabilities. The abstract class HMM provides a method

See also:

transitions to compute the set of Transition's from a given state and a given observation.

We do not have any trellis: at every step we only record the set of possible states with their weight and preceeding state. This structure is called Node. The nodes which have been reached at step t are stored in a map. The keys are states, the values are nodes. The type of states is parametric.

---

Member Function Documentation

vector< State > Tanl::POS::Viterbi::decode	(	double	logtheta,
		State &	startState,
		HMM::Observations &	observations
	)

logtheta is used for beam pruning.

Returns:

the most probable state sequence from startState given the observation sequence observations.

References Tanl::POS::Viterbi::Node::from, and Tanl::POS::Viterbi::Node::weight.

---

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

• tag/Tanlpos/Viterbi.h
• tag/Tanlpos/Viterbi.cpp