benchmarks.md

Benchmarks

Below are the detailed tables with performance measurements for typical word lengths, including alternative implementations from crates.io that have Unicode support.

Due to optimizations, performance of Eddie implementations (except Hamming) changes significantly depending on how different compared strings are, besides their lengths, which is not the case for all the other measured libraries. For this reason Eddie implementations have a pair of measurements, produced under the following scenarios:

1. Comparing the original world to it's version, modified by two random typos.
2. Comparing two unrelated words of the same length (worst case).

Produced on Intel Core i5-4278U 2,6 GHz. All measurements are in nanoseconds and rounded to 2 significant digits.

Levenshtein

	len=3	len=6	len=9	len=12	len=15
eddie 0.3	30 - 50	60 - 130	90 - 260	120 - 410	180 - 610
strsim 0.9	130	210	350	550	760
edit_distance 2.1	160	220	340	490	800
distance 0.4	920	1600	2500	3000	3900

Damerau-Levenshtein

	len=3	len=6	len=9	len=12	len=15
eddie 0.3	220 - 320	310 - 740	370 - 1400	500 - 2400	620 - 3900
strsim 0.9	940	2000	3700	5100	7300
distance 0.4	1800	3200	5400	7700	10900

Hamming

	len=3	len=6	len=9	len=12	len=15
eddie 0.3	12	19	26	36	45
strsim 0.9	12	19	26	36	45
distance 0.4	17	27	36	51	61

Jaro

	len=3	len=6	len=9	len=12	len=15
eddie 0.3	60	90	110 - 130	140 - 190	170 - 230
strsim 0.9	130	170	230	330	410

Jaro-Winkler

	len=3	len=6	len=9	len=12	len=15
eddie 0.3	70	100	110 - 130	140 - 190	170 - 230
strsim 0.9	150	220	240	340	360
natural 0.3	panic	860	1300	1500	1700