WeaviateEx.Query.NearVector (WeaviateEx v0.7.4)

View Source

Near vector query configuration builders.

This module provides functions for building near_vector queries, including support for single vectors, multiple vectors, and per-target named vector configurations.

Single Vector Query

# Standard vector search
vector = [0.1, 0.2, 0.3, ...]
NearVector.single(vector)

List of Vectors Query

# Query with multiple vectors (averaged or combined)
vectors = [vector1, vector2, vector3]
NearVector.list_of_vectors(vectors)

Per-Target Vectors Query

# Different vectors for different named vector spaces
NearVector.per_target(%{
  "title_vector" => title_vec,
  "content_vector" => content_vec
})

These can be used with the Query module:

Query.new("Article")
|> Query.near_vector(NearVector.per_target(%{"title" => vec}))
|> Query.execute(client)

Summary

Types

target_vectors()
vector()

Functions

list_of_vectors(vectors, opts \\ [])

Create a query with a list of vectors.

per_target(targets, opts \\ [])

Create a query with different vectors for different target vector spaces.

single(vector, opts \\ [])

Create a single vector query.

to_api(config)

Convert a near vector configuration to API format.

weighted_targets(weighted_list, opts \\ [])

Create a weighted multi-target vector query.

Types

target_vectors()

@type target_vectors() :: %{required(String.t()) => vector()}

vector()

@type vector() :: [float()]

Functions

list_of_vectors(vectors, opts \\ [])

@spec list_of_vectors(
  [vector()],
  keyword()
) :: map()

Create a query with a list of vectors.

Used for multi-vector search patterns where you want to query with multiple vectors over a single vector space. The server may average or otherwise combine these vectors.

Parameters

  • vectors - A list of vectors (each vector is a list of floats)

Options

  • :certainty - Minimum certainty threshold
  • :distance - Maximum distance threshold

Examples

vectors = [vec1, vec2, vec3]
NearVector.list_of_vectors(vectors)

# With certainty
NearVector.list_of_vectors(vectors, certainty: 0.7)

per_target(targets, opts \\ [])

@spec per_target(
  target_vectors(),
  keyword()
) :: map()

Create a query with different vectors for different target vector spaces.

Used with collections that have multiple named vectors. Allows you to provide a different query vector for each named vector space.

Parameters

  • targets - A map of named vector names to their respective query vectors

Options

  • :certainty - Minimum certainty threshold (applied to all targets)
  • :distance - Maximum distance threshold (applied to all targets)
  • :combination_method - How to combine results (:sum, :average, :minimum, :manual)

Examples

NearVector.per_target(%{
  "title_vector" => title_embedding,
  "content_vector" => content_embedding
})

NearVector.per_target(%{
  "image" => image_vec,
  "text" => text_vec
}, combination_method: :average)

single(vector, opts \\ [])

@spec single(
  vector(),
  keyword()
) :: map()

Create a single vector query.

Parameters

  • vector - A list of floats representing the query vector

Options

  • :certainty - Minimum certainty threshold (0.0 to 1.0)
  • :distance - Maximum distance threshold

Examples

NearVector.single([0.1, 0.2, 0.3])
NearVector.single(vector, certainty: 0.8)

to_api(config)

@spec to_api(map()) :: map()

Convert a near vector configuration to API format.

Examples

config = NearVector.single([0.1, 0.2])
NearVector.to_api(config)
# => %{"vector" => [0.1, 0.2]}

weighted_targets(weighted_list, opts \\ [])

@spec weighted_targets(
  [{String.t(), vector(), float()}],
  keyword()
) :: map()

Create a weighted multi-target vector query.

Allows specifying weights for each target vector space.

Parameters

  • weighted_targets - A list of tuples: {vector_name, vector, weight}

Options

  • :certainty - Minimum certainty threshold
  • :distance - Maximum distance threshold

Examples

NearVector.weighted_targets([
  {"title_vector", title_vec, 0.7},
  {"content_vector", content_vec, 0.3}
])