thefuzz-parity

thefuzz-parity scores string similarity and ranks choices. It counts Unicode scalar values for matcher indices, lengths, and slices. The crate has no runtime dependencies.

The scoring module provides base, partial, token, and weighted scorers. Text processing replaces characters outside Unicode letter and number categories with spaces, lowercases the result, and trims its ends. Interior space runs remain.

use thefuzz_parity::{partial_ratio, token_set_ratio};

assert_eq!(partial_ratio("こんにちは世界", "こんにちは"), 100);
assert_eq!(token_set_ratio("new york", "new york mets", true, true), 100);

The process module ranks borrowed choices and returns each input string unchanged. Cutoffs are inclusive. Equal unrounded scores retain input order.

use thefuzz_parity::process::{extract, Scorer};

let choices = ["new york mets", "chicago cubs", "new york jets"];
let matches = extract("new york", choices, Scorer::WRatio, Some(2), 0);

assert_eq!(matches.len(), 2);
assert_eq!(matches[0].item, "new york mets");

Duplicate removal chooses the longest member of each similarity group. A length tie selects the lexicographically greatest string. The output contains each representative once, ordered by the first input that selects it.

Compatibility

The minimum supported Rust version is 1.70.0. Version 0.1 uses the standard library and forbids unsafe code.

partial_ratio returns the best score over every alignment of the shorter input in the longer one, skipping alignments that cannot beat the best score found so far. Two 1,000-character strings take about 60 microseconds in release mode.

Relation to thefuzz

Not affiliated with SeatGeek or the thefuzz maintainers. Score parity is a tested claim against thefuzz 0.22.1 with rapidfuzz 3.14.5, covering 490 fixed vectors and 2,000 seeded partial_ratio cases with inputs up to 2,000 characters. partial_ratio uses the same pruned window search as rapidfuzz, which thefuzz calls for this scorer.

License

MIT