Hi,

The ListLike instance for ByteString is defined here: https://hackage.haskell.org/package/ListLike-3.1.7.1/docs/src/Data-ListLike-Instances.html

The implementation of Levenshtein distance in language-spelling is here: https://hackage.haskell.org/package/language-spelling-0.3.2/docs/src/Language-Distance.html#levenshtein

It seems to use only O(1) operations on the Bytestring: length and index (!!).
It uses an unboxed mutable vector in the ST monad.
It is specialized for ByteString.

So I think it should be quite efficient already. If you want you can use the same approach to implement the other algorithm, with Data.Bits for the bit-fiddling: https://hackage.haskell.org/package/base-4.8.2.0/docs/Data-Bits.html

Cheers,
Sylvain


On 27/04/2016 01:17, Mason Lai wrote:
Hi,

I'm writing a Haskell program which is looking for whether or not a string (needle) exists within a larger string (haystack) given a Levenshtein distance. Note that it shouldn't calculate the Levenshtein distance between the needle and haystack. For example,

needle: AAA
haystack: TTTTTAATTTTT

The needle would be "found" if the Levenshtein distance is set at 1, because dist("AAT", "AAA") == 1.

The following almost works:

distance(needle, haystack) - (len(haystack) - len(needle))

But it does not handle deletions correctly.

I've previously written a program which does this approximate search in Java with the Wu-Manber extension of the Bitap, or shift-or, algorithm. The same strategy seems difficult to code up in Haskell because it's very "stateful" and involves a lot of bit-fiddling. (Maybe it's actually quite simple, but I'm not sure how I would go about doing this.)

As a further complication, I'd prefer to keep the data packed as ByteStrings, as I'll be dealing with around 200 GiBs of data (split and parallelized over a cluster, but still a lot). I don't really know how to deal with ByteStrings without using the functions that come along in the ByteString module or unpacking the ByteString.

I've been using the language-spelling package, which has a module which can calculate the Levenshtein distance between two ByteStrings. I see that it uses ListLike. I'm not really sure how it works, but I assume it makes ByteString an instance of ListLike, and then you have access to all of the ListLike methods?

Does anyone have any advice on how I should proceed with this? I don't mind learning new things, but I don't really know what the best strategy is or where to look.

-- Mason


_______________________________________________
Beginners mailing list
Beginners@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/beginners