Struct LinearChainCRF 

pub struct LinearChainCRF { /* private fields */ }

Linear-chain CRF for sequence labeling.

This specialized CRF structure enables efficient inference using the forward-backward algorithm and Viterbi decoding.

Implementations

impl LinearChainCRF

pub fn new(num_states: usize) -> Self

Create a new linear-chain CRF.

pub fn add_feature(&mut self, feature: Box<dyn FeatureFunction>, weight: f64)

Add a feature function with its weight.

pub fn set_transition_weights(&mut self, weights: Array2<f64>) -> Result<()>

Set transition weights directly.

This is useful when you have pre-trained weights.

pub fn set_emission_weights(&mut self, weights: Array2<f64>) -> Result<()>

Set emission weights directly.

pub fn viterbi(&self, input_sequence: &[usize]) -> Result<(Vec<usize>, f64)>

Viterbi algorithm for finding the most likely label sequence.

Returns the optimal label sequence and its score.

pub fn forward(&self, input_sequence: &[usize]) -> Result<Array2<f64>>

Forward algorithm for computing marginal probabilities.

Returns forward probabilities: α[t, s] = P(y₁…yₜ = s, x₁…xₜ)

pub fn backward(&self, input_sequence: &[usize]) -> Result<Array2<f64>>

Backward algorithm for computing marginal probabilities.

Returns backward probabilities: β[t, s] = P(yₜ₊₁…yₙ | yₜ = s, xₜ₊₁…xₙ)

pub fn marginals(&self, input_sequence: &[usize]) -> Result<Array2<f64>>

Compute marginal probabilities for each position.

Returns: P(yₜ = s | x) for all t and s

pub fn to_factor_graph(&self, input_sequence: &[usize]) -> Result<FactorGraph>

Convert to factor graph representation.