Vettore v0.3.1

View Source Vettore.Encoding.Muvera (Vettore v0.3.1)

MUVERA-style fixed dimensional encodings for multi-vector retrieval.

This implements the public FDE boundary: query encodings sum vectors in each partition, document encodings average vectors in each partition, and both use the same deterministic partition configuration.

Link to this section Summary

Types

config()
normalized_config()
vector()

Functions

encode_document(vectors, config \\ [])

Encodes document vectors by averaging projected vectors in each partition.

encode_query(vectors, config \\ [])

Encodes query vectors by summing projected vectors in each partition.

Link to this section Types

Link to this type

config()

View Source 
@type config() :: keyword()
Link to this type

normalized_config()

View Source 
@type normalized_config() :: %{
  dimension: pos_integer(),
  num_repetitions: pos_integer(),
  num_simhash_projections: non_neg_integer(),
  seed: non_neg_integer(),
  projection_dimension: pos_integer(),
  final_projection_dimension: pos_integer() | nil
}
Link to this type

vector()

View Source 
@type vector() :: [number()]

Link to this section Functions

Link to this function

encode_document(vectors, config \\ [])

View Source 
@spec encode_document([vector()], config()) :: {:ok, [float()]} | {:error, term()}

Encodes document vectors by averaging projected vectors in each partition.

Link to this function

encode_query(vectors, config \\ [])

View Source 
@spec encode_query([vector()], config()) :: {:ok, [float()]} | {:error, term()}

Encodes query vectors by summing projected vectors in each partition.