Struct HmmModel 

pub struct HmmModel { /* private fields */ }

A trained BMES Hidden Markov Model for segmenting clusters the dictionary has no match for at all.

Ships with no built-in parameters: like with_frequencies, training needs a segmented corpus, and no bundleable, commercially-clean one has been found (see docs/ROADMAP.md Phase 4). Build one with HmmModel::from_counts from BMES tag counts gathered over a corpus you’re licensed to use, then attach it with KhmerTokenizer::with_hmm.

Implementations

impl HmmModel

pub fn from_counts( start_counts: [u64; 4], trans_counts: [[u64; 4]; 4], emit_counts: HashMap<String, [u64; 4]>, ) -> Self

Build a model from raw BMES tag counts: start_counts[state] is how often a Khmer run began tagged state; trans_counts[i][j] is how often a cluster tagged i was immediately followed by one tagged j; emit_counts[cluster][state] is how often cluster was tagged state. All three are add-one smoothed, so an unseen transition or start state is merely unlikely, never impossible.

Trait Implementations

impl Clone for HmmModel

fn clone(&self) -> HmmModel

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.