Struct Vocabulary 

pub struct Vocabulary {
    pub word2idx: HashMap<String, usize>,
    pub idx2word: Vec<String>,
    pub counts: Vec<u64>,
    pub noise_table: Vec<u32>,
    pub total_tokens: u64,
}

Maps tokens ↔ integer indices and stores frequency statistics.

Fields

word2idx: HashMap<String, usize>

word → index

idx2word: Vec<String>

index → word

counts: Vec<u64>

Raw corpus frequency per index

noise_table: Vec<u32>

Flat noise table for O(1) negative sampling

total_tokens: u64

Total token count (after min_count filter)

Implementations

impl Vocabulary

pub fn build(sentences: &[Vec<String>], config: &Config) -> Result<Self>

Build vocabulary from a tokenised corpus.

Steps:

1. Count every token
2. Drop tokens below config.min_count
3. Sort by descending frequency (stable index order)
4. Build unigram noise table

use word2vec::{Config, vocab::Vocabulary};

let corpus = vec!["the cat sat on the mat".to_string()];
let tokens: Vec<Vec<String>> = corpus.iter()
    .map(|s| s.split_whitespace().map(str::to_string).collect())
    .collect();

let vocab = Vocabulary::build(&tokens, &Config::default()).unwrap();
assert!(vocab.word2idx.contains_key("the"));
assert_eq!(vocab.count("the"), 2);

pub fn len(&self) -> usize

Number of unique tokens in vocabulary.

pub fn is_empty(&self) -> bool

Returns true if the vocabulary contains no words.

pub fn count(&self, word: &str) -> u64

Frequency of a word (0 if not in vocab).

use word2vec::{Config, vocab::Vocabulary};
let corpus = vec!["a a b".to_string()];
let tokens: Vec<Vec<String>> = corpus.iter()
    .map(|s| s.split_whitespace().map(str::to_string).collect())
    .collect();
let vocab = Vocabulary::build(&tokens, &Config::default()).unwrap();
assert_eq!(vocab.count("a"), 2);
assert_eq!(vocab.count("z"), 0);

pub fn should_subsample(&self, idx: usize, threshold: f64, dice: f64) -> bool

Returns true if this word should be subsampled (discarded) given a uniformly random dice in [0, 1).

Uses Mikolov’s formula: P(keep) = min(1, sqrt(t/f) + t/f).

pub fn negative_sample(&self, rng: &mut SmallRng) -> usize

Draw a negative sample index from the noise distribution.

Uses the precomputed unigram table for O(1) lookup.

pub fn tokenise_and_subsample( &self, sentence: &[String], threshold: f64, rng: &mut SmallRng, ) -> Vec<usize>

Tokenise and subsample a sentence, returning word indices.

Trait Implementations

impl Clone for Vocabulary

fn clone(&self) -> Vocabulary

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Vocabulary

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Vocabulary

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for Vocabulary

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.