Separately counting the number of deletions in the Levenshtein distance algorithm

Question

So I'm aware that Levenshtein Distance algorithm takes into account the minimum number of deletions, insertions and substitutions required to change a String A into String B. But, I was wondering how you can separately keep track of number of deletions in the total edits required to make the change. I was looking at this implementation of the algorithm,

def levenshtein(first, second)
    first = first.split
    second = second.split
    first_size = first.size
    second_size = second.size
    matrix = [(0..first_size).to_a]
    (1..second_size ).each do |j|
        matrix << [j] + [0] * (first_size)
    end
    count = 0
    (1..second_size).each do |i|
       (1..first_size).each do |j|
         if first[j-1] == second[i-1]
           matrix[i][j] = matrix[i-1][j-1]
         else
           matrix[i][j] = [matrix[i-1][j],matrix[i][j-1], matrix[i-1][j-1]].min + 1
         end
       end
    end
    return matrix.last.last 
end

So in order to keep track of deletions, I tried:

if matrix[i-1[j] == [matrix[i-1][j],matrix[i][j-1], matrix[i-1][j-1]].min

then increase the count. But, this doesn't seem to work. I also tried to get the difference in size for two strings but it fails for the following case

String 1: "my response to prompt#1"
String 2: "my edited response to"

There is clearly 1 deletion here but simply getting the difference in size won't detect so.

I was wondering if anyone knows how to keep track of number of deletions that were involved in the total edits for changing string A into string B.

Deleting a word. For example, `"my response to prompt#1", "my response to"`, we have a deletion of `"prompt#1"` — kchoi, Oct 10 '14 at 08:38
Could you please post a working console example ? It would help going into fixing :) — Sébastien, Oct 10 '14 at 10:17
It's not much of a fix as much as figuring out how to count the number of deletions in the dp process. — kchoi, Oct 10 '14 at 15:54
Ideally, it should pass for the following cases: Case 1: `"my response to prompt#1", "my edited response to prompt#1"` should return 1 because you added `edited.` Case 2: `"haha bob", "haha bobber"` should return 1 because you substituted bob with bobber. Case 3: `"my response to prompt#1", "my response to"` should return -1 because there was 1 deletion. So this is what I want to do, return negative number when it is deleting. — kchoi, Oct 10 '14 at 15:55

score 3 · Accepted Answer · answered Oct 10 '14 at 19:14

We can make the deletion count ride along with the number of substitutions by making each entry of the table a list comprised of the two quantities. (As a side effect, the secondary optimization goal is to minimize the number of deletions. I don't know whether this is desirable or not.)

def levenshtein(first, second)
    first = first.split
    second = second.split
    first_size = first.size
    second_size = second.size
    matrix = [(0..first_size).to_a]
    (1..second_size ).each do |j|
        matrix << [[j,0]] + [[0,0]] * (first_size)
    end
    count = 0
    (1..second_size).each do |i|
       (1..first_size).each do |j|
         if first[j-1] == second[i-1]
           matrix[i][j] = matrix[i-1][j-1]
         else
           matrix[i][j] = [[matrix[i-1][j  ][0]+1, matrix[i-1][j  ][1]  ],
                           [matrix[i  ][j-1][0]+1, matrix[i  ][j-1][1]+1],
                           [matrix[i-1][j-1][0]+1, matrix[i-1][j-1][1]  ]].min
         end
       end
    end
    return matrix.last.last 
end

This seems to fail for `String 1: "the the the" String 2: "the the"` Know why that might be? It should return 1 for matrix.last.last[1] but returns 0 instead. — kchoi, Oct 10 '14 at 20:24
I added `if second_size < first_size && matrix.last.last[1] == 0 @difference = (first_size - second_size) end` and seemed to solve the problem. — kchoi, Oct 10 '14 at 20:41

Separately counting the number of deletions in the Levenshtein distance algorithm

1 Answers1