Crate sketchir 

sketchir: sketching primitives for IR.

This crate is intended for index-only similarity sketches used in:

• near-duplicate detection (MinHash / shingles)
• text fingerprinting (SimHash)
• approximate similarity search (LSH-style candidate generation)

Scope here is primitives: signatures, basic indexing, deterministic behavior. Higher-level workflows (crawl dedupe pipelines, content extraction, etc.) belong elsewhere.

Dense-vector LSH: LSHIndex vs DenseSimHashLSH

Both index dense f32 vectors but differ in lifecycle and recall strategy:

• LSHIndex: batch workflow (add, build, search). Multi-table random projection with tunable num_tables / num_functions for recall control.
• DenseSimHashLSH: incremental insertion (no build step). Single SimHash table with Hamming-distance-1 neighbor probing for approximate recall.

Re-exports

pub use blocking::BlockingConfig;

pub use blocking::MinHashTextLSH;

pub use dense_simhash::DenseSimHashLSH;

pub use lsh::LSHIndex;

pub use lsh::MinHashLSH;

pub use lsh::SimHashLSH;

pub use minhash::MinHash;

pub use minhash::MinHashSignature;

pub use simhash::simhash_fingerprint;

pub use simhash::SimHashFingerprint;

Modules

blocking

Text blocking helpers built on MinHash + LSH.

dense_simhash

Dense-vector SimHash and an embedding LSH index.

lsh

LSH-style indexing helpers.

minhash

MinHash for Jaccard similarity estimation.

simhash

SimHash: binary fingerprints for fast near-duplicate detection.

Enums

Error

Errors for sketchir indexes and operations.