Struct vaporetto::Trainer

pub struct Trainer<'a> { /* private fields */ }

Available on crate feature train only.

Trainer.

Examples

use std::fs::File;
use std::io::{prelude::*, BufReader, BufWriter};

use vaporetto::{Sentence, SolverType, Trainer};

let mut train_sents = vec![];
let f = BufReader::new(File::open("dataset-train.txt").unwrap());
for (i, line) in f.lines().enumerate() {
    train_sents.push(Sentence::from_tokenized(&line.unwrap()).unwrap());
}

let dict: Vec<String> = vec![];
let mut trainer = Trainer::new(3, 3, 3, 3, dict, 0, &[]).unwrap();
for (i, s) in train_sents.iter().enumerate() {
    trainer.add_example(&s);
}

let model = trainer.train(0.01, 1., SolverType::L1RegularizedL2LossSVC).unwrap();
let mut f = BufWriter::new(File::create("model.bin").unwrap());
model.write(&mut f).unwrap();

Implementations

impl<'a> Trainer<'a>

pub fn new( char_window_size: u8, char_ngram_size: u8, type_window_size: u8, type_ngram_size: u8, dict_words: Vec<String>, dict_word_max_len: u8, tag_dictionary: &'a [Sentence<'a, '_>] ) -> Result<Self>

Creates a new trainer.

Arguments

• char_window_size - The character window size.
• char_ngram_size - The character n-gram length.
• type_window_size - The character type window size.
• type_ngram_size - The character type n-gram length.
• dict_words - A word dictionary.
• dict_word_max_len - Dictionary words longer than this value will be grouped together, where the length is in characters.
• tag_dictionary - A tag dictionary. Words not included in the corpus are annotated with the tag specified here.

Errors

If invalid parameters are given, an error variant will be returned.

pub fn add_example<'b>(&mut self, sentence: &'a Sentence<'a, 'b>)

Adds a sentence to the trainer.

pub fn train(self, epsilon: f64, cost: f64, solver: SolverType) -> Result<Model>

Trains word boundaries and tags.

Arguments

• epsilon - The tolerance of the termination criterion.
• cost - The parameter C.
• solver - Solver type.

Errors

If the solver returns an error, that will be propagated.

pub fn n_features(&self) -> usize

Returns the number of boundary features.