Milvex.AnnSearch (milvex v0.10.2)

Represents a single ANN search request for hybrid search.

Supports both vector data (for dense/sparse vector search) and text data (for BM25 full-text search).

Examples

# Vector search
{:ok, search} = AnnSearch.new("embedding", [[0.1, 0.2, 0.3]],
  limit: 10,
  params: %{nprobe: 10}
)

# Text search (BM25)
{:ok, search} = AnnSearch.new("text_sparse", ["search query"],
  limit: 10
)

Summary

Types

query_data()

Query data - either vectors for dense/sparse search or text strings for BM25.

t()

Functions

new(anns_field, data, opts \\ [])

Creates a new ANN search request.

Types

query_data()

@type query_data() :: [[number()]] | [String.t()]

Query data - either vectors for dense/sparse search or text strings for BM25.

t()

@type t() :: %Milvex.AnnSearch{
  anns_field: String.t(),
  data: query_data(),
  expr: String.t() | nil,
  expr_params: map() | nil,
  limit: pos_integer(),
  params: map() | nil
}

Functions

new(anns_field, data, opts \\ [])

@spec new(String.t(), query_data(), keyword()) ::
  {:ok, t()} | {:error, Milvex.Errors.Invalid.t()}

Creates a new ANN search request.

Parameters

anns_field - Name of the vector field to search
data - Query vectors or text strings
opts - Options (see below)

Options

:limit - (required) Maximum results for this sub-search
:params - Search parameters map (e.g., %{nprobe: 10})
:expr - Filter expression string
:expr_params - Template parameters map for the filter expression (e.g., %{"min_year" => 2020})

Examples

{:ok, search} = AnnSearch.new("embedding", [[0.1, 0.2]], limit: 10)
{:ok, search} = AnnSearch.new("text_sparse", ["query"], limit: 5)