Performance Benchmarks

FuzzyBunny is designed for speed, utilizing an optimized C++ core with bit-parallel algorithms and multi-threading support.

Core Matching Performance

The core Levenshtein implementation uses Myers' Bit-Parallel algorithm for strings up to 64 characters. This allows for $O(N)$ matching with very low constant overhead.

For longer strings, it falls back to an optimized $O(NM)$ dynamic programming approach.

Multi-threading with OpenMP

batch_match is parallelized using OpenMP. This allows you to process large numbers of queries across all available CPU cores without being bottlenecked by the Python Global Interpreter Lock (GIL).

In-place Normalization

By pre-normalizing candidates and queries in batch_match, we avoid redundant computations, making it significantly faster for bulk operations compared to repetitive calls to individual matchers.

Benchmarking on Your Machine

You can run the built-in benchmarking tool to see how it performs on your specific hardware and data:

import fuzzybunny

perf = fuzzybunny.benchmark("query", ["candidate"] * 10000)
print(perf)