Trait BaseHiddenMarkovModelSegmenter

Source

pub trait BaseHiddenMarkovModelSegmenter: Sized + Clone {
    // Required methods
    fn hmm(&self) -> &HiddenMarkovModel;
    fn hmm_mut(&mut self) -> &mut HiddenMarkovModel;
    fn from_hmm(hmm: HiddenMarkovModel) -> Self;

    // Provided methods
    fn fit_segmented(&mut self, sents: Vec<Vec<String>>) { ... }
    fn fit_unsegmented(&mut self, sent_strs: Vec<String>) { ... }
    fn save_to_path(&self, path: &str) -> Result<(), ModelError> { ... }
    fn load_from_path(&mut self, path: &str) -> Result<(), ModelError> { ... }
    fn save_to_writer<W: Write>(&self, writer: &mut W) -> Result<(), ModelError> { ... }
    fn load_from_reader<R: Read>(&mut self, reader: R) -> Result<(), ModelError> { ... }
    fn score(&self, sents: Vec<Vec<String>>) -> Result<Vec<f64>, ModelError> { ... }
    fn predict(&self, sent_strs: Vec<String>) -> Vec<Vec<String>> { ... }
    fn predict_with_offsets(
        &self,
        sent_strs: Vec<String>,
    ) -> Vec<Vec<(String, (usize, usize))>> { ... }
}

Expand description

Core HMM segmenter behavior with default implementations.

The segmenter delegates vocabulary building, parameter estimation, Viterbi decoding, and save/load to BaseHiddenMarkovModel. It provides the word↔char↔BMES conversion layer on top.

Required Methods§

Source

fn hmm(&self) -> &HiddenMarkovModel

Source

fn hmm_mut(&mut self) -> &mut HiddenMarkovModel

Source

fn from_hmm(hmm: HiddenMarkovModel) -> Self

Provided Methods§

Source

fn fit_segmented(&mut self, sents: Vec<Vec<String>>)

Train the segmenter from supervised segmented sentences.

Source

fn fit_unsegmented(&mut self, sent_strs: Vec<String>)

Train the segmenter from unsegmented sentences using Baum-Welch EM.

Each string in sent_strs is an unsegmented sentence. Characters are extracted and passed to the underlying HMM’s Baum-Welch algorithm. If the model was previously fitted (e.g., via fit_segmented), the existing parameters serve as EM initialization (warm start).

Source