Rapidfuzz Versions Save

Changed

• significantly improve type hints for the library

Fixed

• fix cmake version parsing

Fixed

• use the correct version of rapidfuzz-cpp when building against a system installed version

Added

• added process.cpdist which allows pairwise comparison of two collection of inputs

Fixed

• fix some minor errors in the type hints
• fix potentially incorrect results of JaroWinkler when using high prefix weights

Changed

• reduce import time

Changed

• upgrade to Cython==3.0.9

Fixed

• upgrade rapidfuzz-cpp which includes a fix for build issues on some compilers
• fix some issues with the sphinx config

Fixed

• fix overflow error on systems with sizeof(size_t) < 8

Fixed

• fix pure python fallback implementation of fuzz.token_set_ratio
• properly link with -latomic if std::atomic<uint64_t> is not natively supported

Performance

• add banded implementation of LCS / Indel. This improves the runtime from O((|s1|/64) * |s2|) to O((score_cutoff/64) * |s2|)

Changed

• upgrade to Cython==3.0.7
• cdist for many metrics now returns a matrix of uint32 instead of int32 by default

Fixed

• use _mm_malloc/_mm_free on macOS if aligned_alloc is unsupported

Fixed

• fix compilation failure on macOS

Changed

• skip pandas pd.NA similar to None
• add score_multiplier argument to process.cdist which allows multiplying the end result scores with a constant factor.
• drop support for Python 3.7

Performance

• improve performance of simd implementation for LCS / Indel / Jaro / JaroWinkler
• improve performance of Jaro and Jaro Winkler for long sequences
• implement process.extract with limit=1 using process.extractOne which can be faster

Fixed

• the preprocessing function was always called through Python due to a broken C-API version check
• fix wraparound issue in simd implementation of Jaro and Jaro Winkler