Crate superkmeans 

Rust port of SuperKMeans — a fast k-means clustering library for high-dimensional vector embeddings using BLAS+ADSampling pruning.

The C++ original lives in SuperKMeans/; this crate re-implements the C++ public API (no Python surface) in pure Rust with no FFI dependencies.

Re-exports

pub use common::DistanceFunction;

pub use common::KnnCandidate;

pub use hierarchical::HierarchicalSuperKMeans;

pub use hierarchical::HierarchicalSuperKMeansConfig;

pub use hierarchical::HierarchicalSuperKMeansIterationStats;

pub use superkmeans::ClusterBalanceStats;

pub use superkmeans::SuperKMeans;

pub use superkmeans::SuperKMeansConfig;

pub use superkmeans::SuperKMeansIterationStats;

pub use utils::TicToc;

pub use utils::compute_l2_squared;

pub use utils::compute_norms_row_major;

pub use utils::find_nearest_neighbor_brute_force;

pub use utils::generate_random_vectors;

pub use utils::make_blobs;

Modules

adsampling

ADSampling-style random rotation + per-dimension pruning thresholds.

batch

Batched nearest-neighbour search built on top of SGEMM and the pruning kernel.

common

Shared constants and small utility types.

distance

Squared-L2 distance with a SIMD-friendly reduction.

gemm

SGEMM backend. Default: pure-Rust matrixmultiply. With a BLAS backend feature the same calls route through cblas_sgemm:

hierarchical

Hierarchical SuperKMeans: 3-phase clustering for very large k.

layout

Centroid layout for the pruning path.

pdxearch

Progressive ADSampling-based top-1 nearest-neighbour search.

superkmeans

Core SuperKMeans algorithm: BLAS+pruning k-means.

utils

Timing, blob data generation, and brute-force reference routines.