Skip to content

Latest commit

 

History

History
62 lines (43 loc) · 3.47 KB

README.md

File metadata and controls

62 lines (43 loc) · 3.47 KB

Narrow Down - Efficient near-duplicate search

PyPI - Version PyPI - Python Version Tests Codecov License

Project Status: Active – The project has reached a stable, usable state and is being actively developed. Black pre-commit Contributor Covenant

Narrow Down offers a flexible but easy-to-use Python API to finding duplicates or similar documents also in very large datasets. It reduces the O(n²) problem of comparing all strings with each other to linear scale by using approximation algorithms like Locality Sensitive Hashing.

Features

  • Document indexing and search based on the Minhash LSH algorithm
  • High performance thanks to a native extension module in Rust
  • Easy-to-use API with automated parameter tuning
  • Works with exchangeable storage backends. Currently implemented:
    • In-Memory
    • Cassandra / ScyllaDB
    • SQLite
    • User defined backends (by implementing a small interface)
  • Native asyncio interface

Installation

The Python package can be installed with pip:

pip install narrow-down

Extras

Some of the heavier functionality is available as extra:

pip install narrow-down[scylladb]   # Cassandra / ScyllaDB storage backend

Similar projects

  • pylsh offers a good implementation of the classic Minhash LSH scheme in Python and Cython. If you only need this and you don't need a database backend it can be a good choice.
  • Datasketch implements an interesting collection of different data sketching algorithms for similarity matching, cardinality estimation and k-nearest-neighbour search. The implementation is not highly optimized but very well usable, the documentation rich and multiple database backends can be used for some of the sketches
  • Milvus offers a full database stack for vector search, a different approach for fast searching. It can also be applied to text search when an embedding like Word2Vec or Bert is used to vectorize the text.

Credits

This package was created with Cookiecutter and the fedejaure/cookiecutter-modern-pypackage project template.