Struct SentenceEncoder 

pub struct SentenceEncoder { /* private fields */ }

Encodes sentences to fixed-length float vectors via word-embedding lookup and pooling.

Words not found in the vocabulary receive an OOV vector (all zeros by default, but they are excluded from mean pooling when all words in the sentence would otherwise be OOV — in that case a zero vector is returned).

Implementations

impl SentenceEncoder

pub fn new(vocab: &[String], config: SentenceEncoderConfig) -> Self

Create a SentenceEncoder with randomly initialised embeddings for every word in vocab.

Embeddings are initialised deterministically from a seeded LCG so that results are reproducible without importing any RNG crate.

pub fn from_vectors( vectors: HashMap<String, Vec<f32>>, config: SentenceEncoderConfig, ) -> Self

Create a SentenceEncoder from a pre-built token-to-vector map.

All vectors must have the same length, which must equal config.embedding_dim. If the map is empty the encoder still works but will return zero vectors for every sentence.

pub fn encode(&self, sentence: &str) -> Vec<f32>

Encode a single sentence to a fixed-length Vec<f32>.

The sentence is split on whitespace (after lower-casing). Tokens beyond max_seq_len are dropped. Words not found in the vocabulary are ignored (treated as if absent) in mean/weighted-mean pooling. For max pooling, missing words contribute a zero vector.

pub fn encode_batch(&self, sentences: &[&str]) -> Vec<Vec<f32>>

Encode a batch of sentences.

pub fn similarity(&self, a: &[f32], b: &[f32]) -> f32

Cosine similarity between two embedding vectors.

Returns a value in [-1.0, 1.0], or 0.0 when either vector has zero norm.

pub fn most_similar<'a>( &self, query: &str, sentences: &[&'a str], top_k: usize, ) -> Vec<(&'a str, f32)>

Find the top_k sentences most similar to query (by cosine similarity), returned in descending similarity order.

pub fn embedding_dim(&self) -> usize

Return the embedding dimensionality.

pub fn embeddings_mut(&mut self) -> &mut HashMap<String, Vec<f32>>

Mutable access to the embeddings map for in-place updates.

Trait Implementations

impl Debug for SentenceEncoder

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

Formats the value using the given formatter.