[TEXT-130] JaroWinklerDistance: Wrong results due to precision of transpositions

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.4
• Fix Version/s: 1.5
• Assignee: Rob Tompkins
• Reporter: Jan Martin Keil
• Created: 02/Aug/18 20:21
• Updated: 18/Oct/18 06:59
• Resolved: 18/Oct/18 06:59

Description

The method JaroWinklerDistance#matches returns transpositions / 2 as integer. However, it is not granted for transpositions to be even. E.g. comparing "aaabcd" and "aaacdb" will result in transpositions = 3. Therefore the method must return 1.5, not 1. Otherwise the similarity is 0.9611111111111111 instead of 0.9416666666666667.

I recommend to return halfTranspositions instead of transpositions and doing the cast and division ((double) mtp[1] / 2) in JaroWinklerDistance#apply.

Issue Links

• links to: GitHub Pull Request #87