Struct SentenceEncoder 

pub struct SentenceEncoder { /* private fields */ }

Projects sequences of token embeddings to a single sentence-level vector.

Internally the encoder applies:

1. Pooling — aggregate token embeddings with the chosen strategy.
2. Projection — a learnable embedding_dim × projection_dim linear layer (bias included) maps the pooled vector to the output space.
3. Optional L2 normalisation to unit length.

Weights are initialised from a deterministic LCG seeded by seed.

Implementations

impl SentenceEncoder

pub fn new( embedding_dim: usize, projection_dim: usize, pooling: PoolingStrategy, seed: u64, ) -> Self

Create a new SentenceEncoder with LCG-initialised weights.

Parameters

• embedding_dim — dimensionality of token embeddings fed to encode.
• projection_dim — output dimensionality of sentence embeddings.
• pooling — pooling strategy.
• seed — deterministic PRNG seed.

pub fn with_normalize(self, normalize: bool) -> Self

Enable or disable L2 normalisation of output embeddings.

pub fn encode(&self, token_embeddings: &[Vec<f64>]) -> Result<Vec<f64>>

Encode a sequence of token embeddings into a single sentence vector.

Returns a Vec<f64> of length projection_dim.

Errors

Returns an error when token_embeddings is empty or any token embedding has a dimension other than embedding_dim.

pub fn cosine_similarity(a: &[f64], b: &[f64]) -> f64

Cosine similarity between two sentence embeddings.

Returns a value in [-1, 1]. Returns 0.0 when either vector has zero norm.

pub fn similarity_matrix( &self, sentences: &[Vec<Vec<f64>>], ) -> Result<Vec<Vec<f64>>>

Encode multiple sentences and return the n × n cosine-similarity matrix.

Each element of sentences is a Vec<Vec<f64>> (token embeddings for one sentence).

Errors

Propagates any error from encode.

pub fn normalize(v: &mut [f64])

L2-normalise a vector in place. A zero-norm vector is left unchanged.

pub fn projection_dim(&self) -> usize

The output (projection) dimension.

pub fn embedding_dim(&self) -> usize

The input (token embedding) dimension.

Trait Implementations

impl Debug for SentenceEncoder

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

Formats the value using the given formatter.