Struct bio::stats::hmm::discrete_emission_opt_end::Model

pub struct Model { /* private fields */ }

Implementation of a hmm::Model with emission values from discrete distributions and an optional declared end state.

Log-scale probabilities are used for numeric stability.

In Rabiner’s tutorial, a discrete emission value HMM has N states and M output symbols. The state transition matrix with dimensions NxN is A, the observation probability distribution is the matrix B with dimensions NxM and the initial state distribution pi has length N. We also included a silent end state ε with vector length N that do not emit symbols for modelling the end of sequences. It’s optional to supply the end probabilities at the creation of the model. If this happens, we’ll create a dummy end state to simulate as if the end state has not been included.

Implementations

impl Model

pub fn new(
    transition: RefCell<Array2<LogProb>>,
    observation: RefCell<Array2<LogProb>>,
    initial: RefCell<Array1<LogProb>>,
    end: RefCell<Array1<LogProb>>,
    has_end_state: bool
) -> Result<Self>

Construct new Hidden MarkovModel with the given transition, observation, and initial state matrices and vectors already in log-probability space.

pub fn with_prob(
    transition: &Array2<Prob>,
    observation: &Array2<Prob>,
    initial: &Array1<Prob>,
    end: Option<&Array1<Prob>>
) -> Result<Self>

Construct new Hidden MarkovModel with the given transition, observation, and initial state matrices and vectors already as Prob values.

pub fn with_float(
    transition: &Array2<f64>,
    observation: &Array2<f64>,
    initial: &Array1<f64>,
    end: Option<&Array1<f64>>
) -> Result<Self>

Construct new Hidden MarkovModel with the given transition, observation, and initial state matrices and vectors with probabilities as f64 values.

Trait Implementations

impl Debug for Model

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Model<usize> for Model

fn num_states(&self) -> usize

The number of states in the model.

fn states(&self) -> StateIter

Return iterator over the states of an HMM.

fn transitions(&self) -> StateTransitionIter

Returns an iterator of all transitions.

fn transition_prob(&self, from: State, to: State) -> LogProb

Transition probability between two states from and to.

fn initial_prob(&self, state: State) -> LogProb

Initial probability given the HMM state.

fn end_prob(&self, _state: State) -> LogProb

End probability given the HMM state.

fn observation_prob(&self, state: State, observation: &usize) -> LogProb

Probability for the given observation in the given state.

fn has_end_state(&self) -> bool

fn transition_prob_idx(&self, from: State, to: State, _to_idx: usize) -> LogProb

Transition probability between two states from and to for observation with index _to_idx (index of to).

impl PartialEq<Model> for Model

fn eq(&self, other: &Model) -> bool

This method tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Model) -> bool

This method tests for !=.

impl Trainable<usize> for Model

fn update_matrices(
    &self,
    transition_hat: Array2<LogProb>,
    observation_hat: Array2<LogProb>,
    initial_hat: Array1<LogProb>,
    end_hat: Array1<LogProb>
)

This feature comes in handy in Bam-Welch algorithm when doing an update of HMM parameters.

fn train_baum_welch(
    &self,
    observations: &[Vec<usize>],
    n_iter: Option<usize>,
    tol: Option<f64>
) -> (Array1<LogProb>, Array2<LogProb>, Array2<LogProb>, Array1<LogProb>)

Iterative procedure to train the model using Baum-Welch algorithm given the training sequences.

impl StructuralPartialEq for Model