Programming

I have searched the internet (including these forums) and perhaps I'm not using the right wording.

What I'm looking for is a function (preferably C) that analyzes the similitude of two numerical or near-numerical values, and returns either a true/false (match/nomatch) or a return code that corresponds to the type of near-match relationship that was encountered. The former would require the user to pre-define (through arguments/environ/variables) the acceptable tolerance or conditions, whereas the latter would be subsequently evaluated by the caller to see if the return code indicated a relationship (between the two numbers) within the tolerances of the application.

For example, each of these could be a different return code and the caller could determine whether or not the relationship is considered a "match" for the purposes of his application:
2401 and 2410 (trailing juxtaposition)
2401 and 4201 (leading juxtaposition)
2401 and 2041 (imbedded juxtaposition)
2479 and 24799 (substring match/trailing dupe)
12 and 12A (this is what I meant above by "near numerical" values)
etc.

Any information that would help me find some code that accomplishes this would be greatly appreciated.
~Marcus

I'm guessing agrep is what you want. agrep == approximate grep

ftp://ftp.cs.arizona.edu/agrep/

I first discovered this concept in PHP, it's called the 'Levenshtein algorithm', and it measures minimum number of changes required to transform one sequence into another. wikipedia has a nice article going over a generic algorithm.

Hi Corona. Thanks for your feedback. We tried LD but it really lacked intelligence for our purposes. For example, the numbers 2401 and 24 it reported as being 50% alike. Clearly by human eye you can see it is and we wouldn't want to include this as a "match." i.e., John Doe at 2401 Main St and John Doe at 24 Main St - probably not the same person. Yet, 2401 and 2410 is a strong possibility. We'd allow this if we did not find John Doe at 2401 Main but we did find a John Doe at 2410, we'd call it a match. Again, LD gives these two numbers a weight of 50% similitude. So in order to catch 2401 and 2410 (simple juxtaposition) we had to set our threshold at 50; yet then we get really crappy matches like 2401 and 24.

In order to catch 310 and 301 with LD, we'd have to lower our threshold to 33%, but then you really let in the junk!

I've no doubt LD has its applications (in fact, we do use it for street names), but it's not so hot for fuzzy number matches. Thank you for your thoughts though.
~Marcus

Perhaps LD could be adjusted by giving a higher cost to adding rather than swapping characters.