Struct ViterbiDecoder 

pub struct ViterbiDecoder {
    pub n_tags: usize,
    pub tag_names: Vec<String>,
}

CRF-style Viterbi decoder operating over emission and transition log-probabilities.

Fields

n_tags: usize

Total number of output tags.

tag_names: Vec<String>

Human-readable tag names in index order.

Implementations

impl ViterbiDecoder

pub fn new(tag_names: Vec<String>) -> Self

Construct a decoder from an ordered list of tag names.

pub fn decode( &self, emissions: &[Vec<f64>], transitions: &[Vec<f64>], ) -> Result<Vec<usize>>

Viterbi decoding over emission scores and a transition matrix.

emissions: [seq_len][n_tags] log-probabilities of each tag at each position. transitions: [n_tags][n_tags] log-probability of transitioning from tag i to tag j.

Returns the most likely tag index sequence.

pub fn indices_to_bio(&self, indices: &[usize]) -> Result<Vec<BioTag>>

Convert a sequence of tag indices to BIO tags.

Tags whose name starts with B- are parsed as BioTag::B(type), I- → BioTag::I(type), O → BioTag::O. Unknown names are treated as O.

pub fn extract_entities(bio_tags: &[BioTag]) -> Vec<(String, usize, usize)>

Extract named entities from a BIO-tagged sequence.

Returns (entity_type, start_index, end_index_exclusive) triples.