fuzzy

String similarity: Levenshtein, Jaro-Winkler, Dice and Jaccard, plus best-match selection over a list. Pure - no capabilities.

ecko get github.com/ecko-lang/fuzzy

import fuzzy

Pure computation: it declares no capabilities, so it cannot touch the network, the filesystem or the environment.

Version 0.10.1 - source - MIT.

fail(message)

The error raised for inputs an algorithm cannot accept.

max2(a, b)

The larger of two numbers.

min2(a, b)

The smaller of two numbers.

levenshtein(a, b)

The Levenshtein distance: the fewest single-character insertions, deletions or substitutions that turn a into b.

Computed with two rows rather than the full matrix, so the memory cost is the length of the shorter string rather than the product of both.

ratio(a, b)

Levenshtein distance normalised to 0.0 (nothing in common) through 1.0 (identical), against the longer of the two strings.

hamming(a, b)

The Hamming distance: how many positions differ. Both strings must be the same length, which is the whole premise of the measure.

jaro(a, b)

The Jaro similarity, 0.0 to 1.0.

Two characters count as matching only if they are the same and fall within max(len_a, len_b) / 2 - 1 positions of each other; a transposition is a matched pair that arrived out of order, and counts half.

jaro_winkler(a, b)

The Jaro-Winkler similarity: Jaro, boosted for strings that share a prefix.

The prefix counts at most four characters, and the boost only applies to already-similar strings. Without the cap a long shared prefix would drag every comparison toward 1.0.

bigrams(s)

The bigrams of a string: every adjacent character pair.

dice(a, b)

The Sorensen-Dice coefficient over bigrams, 0.0 to 1.0. Good for names and titles, where word order varies but letter pairs persist.

jaccard(a, b)

The Jaccard index over character sets: shared characters over distinct characters. Order-blind and the cheapest of the four.

best_match(needle, candidates)

The closest candidate to needle, as { value, score }, or null when there are no candidates. Scored with Jaro-Winkler, which suits the typo-correction case this is usually reached for.

best_match_by(needle, candidates, scorer)

best_match with the scorer of your choice - any fn(a, b) -> float.

top_matches(needle, candidates, limit)

The limit closest candidates, best first, as { value, score } records.

scored(value, score)

One scored candidate. A named function because fn(c) { value: ... } would parse its brace as a block rather than a map literal.

top_matches_by(needle, candidates, limit, scorer)

top_matches with the scorer of your choice.