Enum DistanceMetric 

pub enum DistanceMetric {
    Cosine,
    Euclidean,
    DotProduct,
    Manhattan,
}

Distance metrics supported by HNSW index

Different distance metrics are appropriate for different types of vector data and use cases. The choice affects both search quality and performance.

Variants

• Cosine - Cosine similarity, ideal for normalized vectors and text embeddings
• Euclidean - L2 distance, suitable for general-purpose similarity
• DotProduct - Raw dot product, fast for normalized vectors
• Manhattan - L1 distance, robust to outliers

Performance Characteristics

Metric	SIMD Support	Typical Use Case	Normalization Required
Cosine	Yes	Text embeddings	Recommended
Euclidean	Yes	General similarity	Optional
DotProduct	Yes	Fast approximate cosine	Required
Manhattan	Yes	Robust similarity	Optional

Variants

Cosine

Cosine similarity (1 - normalized dot product) Range: [0, 2] where 0 = identical, 2 = opposite

Euclidean

Euclidean (L2) distance Range: [0, ∞) where 0 = identical

DotProduct

Dot product similarity Range: (-∞, ∞), higher values indicate greater similarity

Manhattan

Manhattan (L1) distance Range: [0, ∞) where 0 = identical

Implementations

impl DistanceMetric

pub fn as_str(&self) -> &str

Get the string representation of this distance metric

Trait Implementations

impl Clone for DistanceMetric

fn clone(&self) -> DistanceMetric

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DistanceMetric

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

Formats the value using the given formatter.

impl Default for DistanceMetric

fn default() -> Self

Returns the “default value” for a type.

impl Display for DistanceMetric

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

Formats the value using the given formatter.

impl PartialEq for DistanceMetric

fn eq(&self, other: &DistanceMetric) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for DistanceMetric

impl Eq for DistanceMetric

impl StructuralPartialEq for DistanceMetric