Struct ExactKnn 

pub struct ExactKnn { /* private fields */ }

Exact k-NN search operator using brute force.

Computes distances to all provided vectors and returns the K nearest. This is useful for:

• Small datasets where HNSW overhead isn’t justified
• Validating HNSW results
• Post-filtering a small candidate set from graph traversal

Complexity

O(n * d) where n is the number of vectors and d is the dimension. For large datasets, use AnnScan instead.

Example

use manifoldb_vector::ops::{ExactKnn, VectorOperator, SearchConfig};
use manifoldb_vector::distance::DistanceMetric;

let vectors = vec![
    (EntityId::new(1), embedding1),
    (EntityId::new(2), embedding2),
];

let config = SearchConfig::k_nearest(5);
let mut knn = ExactKnn::new(vectors, query, DistanceMetric::Cosine, config)?;

while let Some(m) = knn.next()? {
    println!("Entity {:?} at distance {}", m.entity_id, m.distance);
}

Implementations

impl ExactKnn

pub fn new<I>( vectors: I, query: &Embedding, metric: DistanceMetric, config: SearchConfig, ) -> Result<Self, VectorError>

where I: IntoIterator<Item = (EntityId, Embedding)>,

Create a new exact k-NN search operator.

Arguments

• vectors - Iterator of (entity_id, embedding) pairs to search
• query - The query embedding
• metric - Distance metric to use
• config - Search configuration

Errors

Returns an error if any vector has a dimension mismatch with the query.

pub fn k_nearest<I>( vectors: I, query: &Embedding, metric: DistanceMetric, k: usize, ) -> Result<Self, VectorError>

where I: IntoIterator<Item = (EntityId, Embedding)>,

Create a k-nearest search from an iterator of vectors.

Convenience method for common case.

pub fn within_distance<I>( vectors: I, query: &Embedding, metric: DistanceMetric, max_distance: f32, ) -> Result<Self, VectorError>

where I: IntoIterator<Item = (EntityId, Embedding)>,

Create a within-distance search from an iterator of vectors.

Returns all vectors within the specified distance threshold.

pub fn from_slice( vectors: &[(EntityId, Embedding)], query: &Embedding, metric: DistanceMetric, config: SearchConfig, ) -> Result<Self, VectorError>

Create from a slice of vectors (borrows and clones).

Useful when you have a reference to existing data.

pub fn len(&self) -> usize

Get the number of results found.

pub fn is_empty(&self) -> bool

Check if no results were found.

pub fn peek(&self) -> Option<&VectorMatch>

Peek at the next result without consuming it.

pub fn reset(&mut self)

Reset the iterator to the beginning.

pub fn as_slice(&self) -> &[VectorMatch]

Get all results as a slice.

Trait Implementations

impl VectorOperator for ExactKnn

fn next(&mut self) -> Result<Option<VectorMatch>, VectorError>

Get the next match from the operator.

fn dimension(&self) -> usize

Get the dimension of vectors this operator works with.

fn collect_all(&mut self) -> Result<Vec<VectorMatch>, VectorError>

Collect all remaining matches into a vector.