Struct HiddenMarkovModel 

pub struct HiddenMarkovModel { /* private fields */ }

Hidden Markov Model builder.

A temporal model with hidden states and observations.

Implementations

impl HiddenMarkovModel

pub fn new( num_states: usize, num_observations: usize, time_steps: usize, ) -> Self

Create a new HMM.

Arguments

• num_states - Number of hidden states
• num_observations - Number of observable symbols
• time_steps - Length of sequence

pub fn set_initial_distribution( &mut self, initial: ArrayD<f64>, ) -> Result<&mut Self>

Set initial state distribution P(state_0).

pub fn set_transition_matrix( &mut self, transition: ArrayD<f64>, ) -> Result<&mut Self>

Set transition matrix P(state_t | state_{t-1}).

Arguments

• transition - Transition probabilities [from_state, to_state]

pub fn set_emission_matrix( &mut self, emission: ArrayD<f64>, ) -> Result<&mut Self>

Set emission matrix P(obs_t | state_t).

Arguments

• emission - Emission probabilities [state, observation]

pub fn graph(&self) -> &FactorGraph

Get the underlying factor graph.

pub fn filter(&self, observations: &[usize], t: usize) -> Result<ArrayD<f64>>

Perform filtering: compute P(state_t | obs_0:t).

Uses variable elimination to compute the marginal distribution over the hidden state at time t given observations from 0 to t.

pub fn smooth(&self, observations: &[usize], t: usize) -> Result<ArrayD<f64>>

Perform smoothing: compute P(state_t | obs_0:T).

Uses variable elimination with all observations to compute the marginal distribution over the hidden state at time t.

pub fn viterbi(&self, observations: &[usize]) -> Result<Vec<usize>>

Compute most likely state sequence (Viterbi algorithm).

Finds the most probable sequence of hidden states given observations using dynamic programming.