Struct hnsw::DiscreteHNSW

pub struct DiscreteHNSW<T: DiscreteDistance, M: ArrayLength<u32> = U12, M0: ArrayLength<u32> = U24, R = Pcg64> { /* fields omitted */ }

This provides a HNSW implementation for any discrete distance function.

This can achieve much higher performance than a non-discrete distance function due to the discrete distance priority queue used to search. This can be combined with the Hamming struct to add binary features to the HNSW.

Methods

impl<T: DiscreteDistance, M: ArrayLength<u32>, M0: ArrayLength<u32>, R> DiscreteHNSW<T, M, M0, R> where
    R: RngCore + SeedableRng, 

pub fn new() -> Self

Creates a new HNSW with a PRNG which is default seeded to produce deterministic behavior.

pub fn new_params(params: Params) -> Self

Creates a new HNSW with a default seeded PRNG and with the specified params.

impl<T: DiscreteDistance, M: ArrayLength<u32>, M0: ArrayLength<u32>, R> DiscreteHNSW<T, M, M0, R> where
    R: RngCore, 

pub fn new_prng(prng: R) -> Self

Creates a HNSW with the passed prng.

pub fn new_params_and_prng(params: Params, prng: R) -> Self

Creates a HNSW with the passed params and prng.

pub fn insert(&mut self, q: T, searcher: &mut DiscreteSearcher<T>) -> u32

Inserts a feature into the HNSW.

Important traits for &'_ [u8]

impl<'_> Read for &'_ [u8]impl<'_> Write for &'_ mut [u8]

pub fn nearest<'a>(
    &self,
    q: &T,
    ef: usize,
    searcher: &mut DiscreteSearcher<T>,
    dest: &'a mut [u32]
) -> &'a mut [u32]

Does a k-NN search where q is the query element and it attempts to put up to M nearest neighbors into dest. ef is the candidate pool size. ef can be increased to get better recall at the expense of speed. If ef is less than dest.len() then dest will only be filled with ef elements.

Returns a slice of the filled neighbors.

pub fn feature(&self, item: u32) -> &T

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

Trait Implementations

impl<T: DiscreteDistance, M: ArrayLength<u32>, M0: ArrayLength<u32>, R> Default for DiscreteHNSW<T, M, M0, R> where
    R: SeedableRng, 

fn default() -> Self

Returns the "default value" for a type.

impl<T: Clone + DiscreteDistance, M: Clone + ArrayLength<u32>, M0: Clone + ArrayLength<u32>, R: Clone> Clone for DiscreteHNSW<T, M, M0, R>

fn clone(&self) -> DiscreteHNSW<T, M, M0, R>

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.