The dense spatial family covers sqeuclidean, euclidean, and angular.
The main value over naive loops is the combination of SIMD and safer accumulation policy.
use numkong::Euclidean;
let a = [1_i8, 2, 3, 4];
let b = [4_i8, 3, 2, 1];
let distance = i8::euclidean(&a, &b).unwrap(); // widened output, not int8
println!("{distance}");AngularComputes the angular distance, also called cosine distance, between two vectors.3pub trait Angular: StorageElementd = 1 − (a · b) / (‖a‖ × ‖b‖)
Range: [0, 2]. Fails with Error::ShapeMismatch if lengths differ, or
Error::KernelFailed carrying the kernel's status.
Implemented for: f64, f32, f16, bf16, i8, u8, e4m3, e5m2, e2m3, e3m2,
i4x2, and u4x2.
OutputResult type of the angular distance.type Output;angularComputes the angular distance between two equally long slices.fn angular(a: &[Self], b: &[Self]) -> Result<Self::Output, Error>cosineAlias for angular.fn cosine(a: &[Self], b: &[Self]) -> Result<Self::Output, Error>EuclideanComputes the Euclidean distance, the L2 norm of the difference, between two vectors.4pub trait Euclidean: StorageElementd = √(∑ᵢ (aᵢ − bᵢ)²)
Range: [0, ∞). Fails with [Error::ShapeMismatch] if lengths differ, or
Error::KernelFailed carrying the kernel's status.
Implemented for: f64, f32, f16, bf16, i8, u8, e4m3, e5m2, e2m3, e3m2,
i4x2, and u4x2.
SqEuclideanOutputResult type of the squared Euclidean distance.type SqEuclideanOutput;EuclideanOutputResult type of the Euclidean distance.type EuclideanOutput;sqeuclideanSquared Euclidean distance, i.e.fn sqeuclidean(a: &[Self], b: &[Self]) -> Result<Self::SqEuclideanOutput, Error>L2². Faster than euclidean for comparisons.
euclideanEuclidean distance, i.e.fn euclidean(a: &[Self], b: &[Self]) -> Result<Self::EuclideanOutput, Error>L2. True metric distance.
SpatialSimilaritySpatialSimilarity bundles spatial distance metrics: Dot, Angular, and Euclidean.pub trait SpatialSimilarity: Dot + Angular + Euclideanimpl<Scalar: Dot + Angular + Euclidean> SpatialSimilarity for Scalar