Struct AnnIndex 

pub struct AnnIndex { /* private fields */ }

A built RP-forest index over L2-normalized vectors.

Implementations

impl AnnIndex

pub fn build(data: &[Vec<f64>], config: &AnnConfig) -> Self

Build the index from raw (un-normalized) vectors. Each vector is L2-normalized internally. All vectors must have the same dimensionality. Panics if data is empty or dimensions disagree.

pub fn build_normalized(normalized: Vec<Vec<f64>>, config: &AnnConfig) -> Self

Build from pre-normalized vectors (avoids a redundant normalization pass when the caller already has unit vectors). Enforces the same contract as Self::build: panics if normalized is empty or dimensions disagree.

pub fn query(&self, query: &[f64], k: usize) -> Vec<(usize, f64)>

Find the k approximate nearest neighbors of query by cosine similarity. Returns (index, similarity) pairs sorted by descending similarity. query is L2-normalized internally.

pub fn query_by_index(&self, index: usize, k: usize) -> Vec<(usize, f64)>

Find the k approximate nearest neighbors of the item at index (excludes self from results).

pub fn knn_graph(&self, k: usize) -> Vec<Vec<usize>>

Build a full k-NN adjacency list for all items. Returns knn[i] = indices of the k nearest neighbors of item i (excluding self), sorted by descending similarity.

pub fn knn_graph_with_sims(&self, k: usize) -> Vec<Vec<(usize, f64)>>

Like Self::knn_graph, but keeps each neighbor’s cosine similarity: graph[i] = (index, similarity) pairs for the k nearest neighbors of item i (excluding self), sorted by descending similarity.

pub fn len(&self) -> usize

Number of indexed items.

pub fn is_empty(&self) -> bool

True if the index contains no items.