Comments on: Finding the Levenshtein Distance in Scala http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/ The weblog of Tony Morris Tue, 06 Jan 2009 13:36:38 +0000 http://wordpress.org/?v=2.6 By: Anonymous Coward http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/#comment-2280 Anonymous Coward Thu, 08 May 2008 00:36:24 +0000 http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/#comment-2280 Gross... Preview is fin and posts are fuxxored : third try. That blog/post response is bugged... Preview is fine and it screws the posts once you click on 'submit comment'. What you see is NOT what you get :) Would this work : <code> if(a &#38;lt b) a else if(b &#38;lt= c) b else m(f)(i)(mx(i, _))(j - 1) + 1 </code> Gross… Preview is fin and posts are fuxxored : third try. That blog/post response is bugged… Preview is fine and it screws the posts once you click on ’submit comment’. What you see is NOT what you get :)

Would this work :


if(a &#38;lt b) a else if(b &#38;lt= c) b else m(f)(i)(mx(i, _))(j - 1) + 1

]]> By: Anonymous Coward http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/#comment-2279 Anonymous Coward Thu, 08 May 2008 00:34:18 +0000 http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/#comment-2279 erf... Post got fuxxored. I meant could the two lines be rewritten to this : if(a <b> erf… Post got fuxxored.

I meant could the two lines be rewritten to this :

if(a

]]> By: Anonymous Coward http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/#comment-2278 Anonymous Coward Thu, 08 May 2008 00:32:25 +0000 http://blog.tmorris.net/finding-the-levenshtein-distance-in-scala/#comment-2278 Hi, I wrote a spellchecker years ago : I used lots of differents algos (including double-metaphone and algo finding 'probable typos') to find 'candidates', then I was using a Levenshtein edit distance to sort the candidates from 'most relevant' to 'less relevant'. Anyway... I've got questions : first, when working with large strings, using O(mn) memory, Chas Emerick noticed that he could not compare huge strings without getting out-of-memory errors, which lead him to write a version more efficient memory-wise (maybe he's doing DNA string matching ? No idea...). Anyway here is his algo: http://www.merriampark.com/ldjava.htm Does your Scala version here use O(mn) memory ? If so, can it easily be changed to use less memory ? You write: <i>the code builds the call stack as it traverses the matrix (the loopy one does not)</i>. So your version is actually a recursive version right? If so then I guess it's unsuitable for big strings (even if the O(mn) memory problem, should it be in your version, could be fixed)!? Then you write : <i>The code has a better complexity than the typical loopy version by using lazy evaluation (notice that ‘c’ is not always evaluated)</i> I'm not familiar with Scala and I'm curious about it. Could the following : <code> lazy val c = m(f)(i)(mx(i, _))(j - 1) + 1 if(a <b> be rewritten with : <code> if(a <b> and... Would that change anything (Scala-wise) ? In my view, you used what is called the 'Top-down' approach to implement the DP algo: memoization and recursion combined. While the more memory-efficient and more stack-efficient approach by Chas Emerick is the 'Bottom-up' approach ('Top-down' and 'Bottom-up' are defined on Wikipedia, under the Dynamic programming entry). I must admit I'm surprised by the definition on Wikipedia : to me the 'top-down' approach weren't DP, but simply memoization + recursion. Anectodically, I find the bottom-up DP version of this algo easier to read then the recursive+memoization version (I mean, even if you were to write a recursive+memoization version of this DP algo in Java ;) Lastly, after 0-1 Knapsak and edit distance, do you plan to have fun with all the DP algos ? (Floyd-Warshall being next !? ;) Hi,

I wrote a spellchecker years ago : I used lots of differents algos (including double-metaphone and algo finding ‘probable typos’) to find ‘candidates’, then I was using a Levenshtein edit distance to sort the candidates from ‘most relevant’ to ‘less relevant’. Anyway…

I’ve got questions : first, when working with large strings, using O(mn) memory, Chas Emerick noticed that he could not compare huge strings without getting out-of-memory errors, which lead him to write a version more efficient memory-wise (maybe he’s doing DNA string matching ? No idea…). Anyway here is his algo:

http://www.merriampark.com/ldjava.htm

Does your Scala version here use O(mn) memory ? If so, can it easily be changed to use less memory ?

You write: the code builds the call stack as it traverses the matrix (the loopy one does not). So your version is actually a recursive version right? If so then I guess it’s unsuitable for big strings (even if the O(mn) memory problem, should it be in your version, could be fixed)!?

Then you write :

The code has a better complexity than the typical loopy version by using lazy evaluation (notice that ‘c’ is not always evaluated)

I’m not familiar with Scala and I’m curious about it. Could the following :


lazy val c = m(f)(i)(mx(i, _))(j - 1) + 1
if(a

be rewritten with :


if(a

and… Would that change anything (Scala-wise) ?

In my view, you used what is called the ‘Top-down’ approach to implement the DP algo: memoization and recursion combined. While the more memory-efficient and more stack-efficient approach by Chas Emerick is the ‘Bottom-up’ approach (’Top-down’ and ‘Bottom-up’ are defined on Wikipedia, under the Dynamic programming entry).

I must admit I’m surprised by the definition on Wikipedia : to me the ‘top-down’ approach weren’t DP, but simply memoization + recursion.

Anectodically, I find the bottom-up DP version of this algo easier to read then the recursive+memoization version (I mean, even if you were to write a recursive+memoization version of this DP algo in Java ;)
Lastly, after 0-1 Knapsak and edit distance, do you plan to have fun with all the DP algos ? (Floyd-Warshall being next !? ;)

]]>