Comments on: Comparing Strings: An Analysis of Diff Algorithms http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/ A blog for odd things and odd thoughts. Wed, 10 Mar 2010 17:00:21 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Maarten http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-223014 Maarten Tue, 23 Feb 2010 17:41:10 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-223014 The "Myers86" reference is most likely "<a href="http://en.scientificcommons.org/42915130" rel="nofollow">An O(ND) Difference Algorithm and Its Variations</a>", published in 1986 by Eugene W. Myers. The “Myers86″ reference is most likely “An O(ND) Difference Algorithm and Its Variations“, published in 1986 by Eugene W. Myers.

]]> By: aj http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-44432 aj Sat, 16 Jun 2007 15:00:49 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-44432 Hi, How about a minor modification to the diff heuristic such that when one file is a proper subset of the other (meaning on only deletions required), it would, without reducing the efficiency give the best diff? Please let me know your thoughts on this. Regards, Arjun Hi,

How about a minor modification to the diff heuristic such that when one file is a proper subset of the other (meaning on only deletions required), it would, without reducing the efficiency give the best diff?

Please let me know your thoughts on this.

Regards,
Arjun

]]> By: entY8 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-40925 entY8 Wed, 23 May 2007 20:07:38 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-40925 Thanks a lot for this; it was the first useful site I found on this topic :-D (btw, I found it via Google somewhere among the first few hits) Thanks a lot for this; it was the first useful site I found on this topic :-D
(btw, I found it via Google somewhere among the first few hits)

]]> By: turbo http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-8885 turbo Sat, 01 Jul 2006 14:31:07 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-8885 I have been recently speaking to Dr. Gene Myers. His algorithms are very elegant and fast, however as I found out, and as you clearly realized, they do not allow for substitutions, which is one of the allowed operations of Levenshtein distance. I have coded up a few versions of code that computes Levenshtein distance. I wrote the really obvious one, as well as the one dimensional storage version of that. I acutally used that code for many years. After some gentle encouragement, I searched the internet for better algorithms. I found references to Ukkonen's work. I optimized my code to only compute the diagonals needed. That sped up my code by roughly a factor of 6. I knew of his other optimization, which computes the cells of the 2d array in order of distance. I also coded that up, however I have not found an efficient way to do so. There are several operations in my code that are O(n^2), which is unfortunate. The algorithm does fill in less values in the 2d array, but it takes much too long. Perhaps I will figure out some very clever, efficient way to fill out the distances of the array in order... I also coded up a classic implementation of the Damerau/Levenshtein distance which also allows for transpositition. Thanks for fixing the typo in my array where I included an extra value... If you have access to Ukkonen's papers, I would really like to see them. They seem to be too old to be online.... turbo I have been recently speaking to Dr. Gene Myers. His algorithms are very elegant and fast, however as I found out, and as you clearly realized, they do not allow for substitutions, which is one of the allowed operations of Levenshtein distance.

I have coded up a few versions of code that computes Levenshtein
distance. I wrote the really obvious one, as well as the one dimensional storage version of that. I acutally used that code for many years.

After some gentle encouragement, I searched the internet for better algorithms. I found references to Ukkonen’s work. I optimized my code to only compute the diagonals needed. That sped up my code by roughly a factor of 6. I knew of his other optimization, which computes the cells of the 2d array in order of distance. I also coded that up, however I have not found an efficient way to do so. There are several operations in my code that are O(n^2), which is unfortunate. The algorithm does fill in less values in the 2d array, but it takes much too long. Perhaps I will figure out some very clever, efficient way
to fill out the distances of the array in order…

I also coded up a classic implementation of the Damerau/Levenshtein distance which also allows for transpositition.

Thanks for fixing the typo in my array where I included an extra
value…

If you have access to Ukkonen’s papers, I would really like to see them. They seem to be too old to be online….

turbo

]]> By: Julian http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-8652 Julian Tue, 27 Jun 2006 22:08:15 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-8652 Thanks Turbo, You made me very happy, because your comment was evidence that someone actually read - and understood - the report! In your grid, it appears that the edit distance between the string <code>c</code> and <code>h</code> is 1. This is only true if you permit the Replace operation (i.e. Replace 'c' with 'h'.) In the analysis of that section, it assumed the only basic operations were Add and Delete (i.e. Delete 'c'. Add 'h'.) I agree that, if you can use Replace as one of the basic operations, then the increment going down the diagonal can either be 0 or 1 (not 2), and the rest of the analysis needs to be appropriately re-jiggered to cover that. Thanks Turbo,

You made me very happy, because your comment was evidence that someone actually read – and understood – the report!

In your grid, it appears that the edit distance between the string c and h is 1.

This is only true if you permit the Replace operation (i.e. Replace ‘c’ with ‘h’.) In the analysis of that section, it assumed the only basic operations were Add and Delete (i.e. Delete ‘c’. Add ‘h’.)

I agree that, if you can use Replace as one of the basic operations, then the increment going down the diagonal can either be 0 or 1 (not 2), and the rest of the analysis needs to be appropriately re-jiggered to cover that.

]]> By: turbo http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-8580 turbo Tue, 27 Jun 2006 13:11:17 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-8580 Nice paper. However, in the O((|u|+|v|).d) Algorithm you state the increment going down a diagonal is either 0 or 2, which is incorrect. A simple counter example is the array for the string 'cat' and 'hat' which is : <code> xxcat x0123 h1123 a2212 t3321 </code> Since the edit distance is 1, and the upper left corner is 0, this shows that the increment can be 1. In fact, any pair of strings with edit distance of 1 will exhibit this property. Nice paper. However, in the O((|u|+|v|).d) Algorithm you state the increment going down a diagonal is either 0 or 2, which is incorrect. A simple counter example is the array for the string ‘cat’ and ‘hat’
which is :

xxcat
x0123
h1123
a2212
t3321

Since the edit distance is 1, and the upper left corner is 0, this shows that the increment can be 1. In fact, any pair of strings with edit distance of 1 will exhibit this property.

]]> By: girtby.net » Blog Archive » Responding to Adrian http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/comment-page-1/#comment-7167 girtby.net » Blog Archive » Responding to Adrian Mon, 05 Jun 2006 12:14:09 +0000 http://www.somethinkodd.com/oddthinking/2006/01/16/comparing-strings-an-analysis-of-diff-algorithms/#comment-7167 [...] For a good introduction to diffing and the algorithms commonly used, see OddThinking. [...] [...] For a good introduction to diffing and the algorithms commonly used, see OddThinking. [...]

]]>