Struct BinaryVector 

pub struct BinaryVector { /* private fields */ }

Binary quantized vector stored as packed u64 words

Implementations

impl BinaryVector

pub fn from_f32(values: &[f32]) -> BinaryVector

Create a binary vector from fp32 values using sign-based quantization

Positive values become 1, negative/zero become 0

pub fn from_f32_threshold(values: &[f32], threshold: f32) -> BinaryVector

Create a binary vector from threshold-based quantization

Values above threshold become 1, below become 0

pub fn from_f32_median(values: &[f32]) -> BinaryVector

Create a binary vector from median-based quantization

Values above median become 1, below become 0. Better for non-normalized vectors.

pub fn from_raw(data: Vec<u64>, dim: usize) -> BinaryVector

Create from raw packed data

pub fn dim(&self) -> usize

Get the dimensionality of the original vector

pub fn data(&self) -> &[u64]

Get the packed binary data

pub fn size_bytes(&self) -> usize

Get size in bytes

pub fn hamming_distance(&self, other: &BinaryVector) -> u32

Compute Hamming distance to another binary vector

Hamming distance = number of positions where bits differ. Uses popcount which is a single CPU instruction on modern x86.

pub fn hamming_distance_normalized(&self, other: &BinaryVector) -> f32

Compute normalized Hamming distance (0.0 to 1.0)

0.0 = identical, 1.0 = completely different

pub fn approx_cosine_similarity(&self, other: &BinaryVector) -> f32

Convert Hamming distance to approximate cosine similarity

For normalized embeddings, there’s a relationship between Hamming distance and cosine similarity: cos_sim ≈ 1 - 2 * (hamming_dist / dim)

Trait Implementations

impl Clone for BinaryVector

fn clone(&self) -> BinaryVector

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BinaryVector

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.