Struct FDEEncoder 

pub struct FDEEncoder<T: FDEFloat> {
    pub buckets: usize,
    pub dim: usize,
    pub hyperplanes: Array2<T>,
}

Fixed Dimensional Encoder (FDE) implementation.

Encodes variable-length token embeddings into fixed-length vectors using randomized hyperplanes and aggregation.

Fields

• buckets: Number of hyperplanes / buckets to hash tokens into.
• dim: Embedding dimensionality of input tokens.
• hyperplanes: Random hyperplanes matrix used for projection and hashing.

Fields

buckets: usize

dim: usize

hyperplanes: Array2<T>

Implementations

impl<T: FDEFloat> FDEEncoder<T>

pub fn new(buckets: usize, dim: usize, seed: u64) -> Self

Creates a new FDE encoder with the specified number of buckets and embedding dimension.

Arguments

• buckets: Number of hash buckets (hyperplanes).
• dim: Dimensionality of token embeddings.
• seed: RNG seed for reproducible hyperplane initialization.

Returns

A new FDEEncoder instance.

Trait Implementations

impl Default for FDEEncoder<f32>

fn default() -> Self

Returns the “default value” for a type.

impl<T: FDEFloat + Send + Sync + FromPrimitive> FDEEncoding<T> for FDEEncoder<T>

fn encode( &self, multi_vector_tokens: ArrayView2<'_, T>, mode: Aggregation, ) -> Array1<T>

Encode a single multi-vector of tokens into a fixed-dimensional vector.

Projects tokens onto hyperplanes, hashes them into buckets, aggregates by sum or average per bucket, and concatenates the results.

Arguments

• multi_vector_tokens: 2D array of token embeddings (num_tokens, dim).
• mode: Aggregation mode (Sum or Avg).

Returns

A 1D array of length buckets * dim representing the encoded vector.

fn batch_encode( &self, batch_tokens: ArrayView3<'_, T>, mode: Aggregation, ) -> Array2<T>

where T: FDEFloat + Sync + Send, Self: Sync,

Encode a batch of multi-vectors using parallel processing.

Divides the batch across threads for concurrent encoding.

Arguments

• batch_tokens: 3D array (batch_size, num_tokens, dim).
• mode: Aggregation mode (Sum or Avg).

Returns

2D array of encoded vectors (batch_size, buckets * dim).

fn encode_query(&self, tokens: ArrayView2<'_, T>) -> Array1<T>

Encode a query token embedding using sum aggregation.

fn encode_doc(&self, tokens: ArrayView2<'_, T>) -> Array1<T>

Encode a document token embedding using average aggregation.

fn encode_query_batch(&self, batch_tokens: ArrayView3<'_, T>) -> Array2<T>

Batch encode queries using sum aggregation.

fn encode_doc_batch(&self, batch_tokens: ArrayView3<'_, T>) -> Array2<T>

Batch encode documents using average aggregation.