Fast and generic distance functions for high-dimensional data.

Add this to your project:

> cargo add [email protected]

Use it in your project:

use distances::Number;
use distances::vectors::euclidean;

let a = [1.0_f32, 2.0, 3.0];
let b = [4.0_f32, 5.0, 6.0];

let distance: f32 = euclidean(&a, &b);

assert!((distance - (27.0_f32).sqrt()).abs() < 1e-6);

• A Number trait to abstract over different numeric types.
  • Distance functions are generic over the return type implementing Number.
  • Distance functions may also be generic over the input type being a collection of Numbers.
• SIMD accelerated implementations for float types.
• Python bindings with maturin and pyo3.
• no_std support.

• Vectors (high-dimensional data):
  • euclidean
  • squared_euclidean
  • manhattan
  • chebyshev
  • minkowski
    • General Lp-norm.
  • minkowski_p
    • General Lp-norm to the pth power.
  • cosine
  • hamming
  • canberra
    • Canberra Distance
  • bray_curtis
    • Bray-Curtis Distance
  • pearson
    • 1.0 - r where r is the Pearson Correlation Coefficient
• Probability distributions:
  • wasserstein
    • Wasserstein Distance
  • bhattacharyya
    • Bhattacharyya Distance
  • hellinger
    • Hellinger Distance
• String data, e.g. for genomic sequences:
  • levenshtein
  • needleman_wunsch
  • smith_waterman
  • hamming
  • Normalized versions of the above.
• Sets:
  • jaccard
  • hausdorff
    • Hausdorff Distance
• Graphs:
  • tanamoto
• Time series:
  • dtw
    • Dynamic Time Warping
  • msm
    • Move-Split-Merge
  • erp
    • Edit distance with Real Penalty

Contributions are welcome, encouraged, and appreciated! See CONTRIBUTING.md.

Licensed under the MIT license.