Crate diskann_rs

DiskAnnRS

A DiskANN-like Rust library implementing approximate nearest neighbor search with single-file storage support. The library provides both Euclidean distance and Cosine similarity metrics, with a three-layer hierarchical search structure.

Features

• Single-file storage format
• Support for both Euclidean and Cosine distance metrics
• Parallel index construction using Rayon
• Memory-mapped file access for efficient searches
• Three-layer hierarchical search structure

Example

use diskann_rs::{SingleFileDiskANN, DistanceMetric};

// Build a new index
let index = SingleFileDiskANN::build_index_singlefile(
    1000,    // number of vectors
    128,     // dimensionality
    32,      // maximum degree
    0.1,     // fraction of vectors in top layer
    0.2,     // fraction of vectors in middle layer
    DistanceMetric::Euclidean,
    "index.db"
).unwrap();

// Search the index
let query = vec![0.0; 128];  // your query vector
let neighbors = index.search(&query, 10, 64);  // find top 10 with beam width 64

Structs

SingleFileDiskANN

Main struct representing a DiskANN index

Enums

DiskAnnError

Custom error type for DiskAnnRS operations

DistanceMetric

Supported distance metrics for vector comparison