Can AILevenshteinDistance be applied to numbers?

[jordanmontt]

Does the levenshtein distance can be applied to Arrays with numbers? Like #(1 2 3) distanceTo: #(3 4 5) using: AILevenshteinDistance

Because from what I have been reading this distance algorithm is only applied to Strings.

Originally posted by @jordanmontt in #20

[Alokzh]

Hi @jordanmontt, I just tested the current implemntation with array of numbrs & found that it work the same way as it works in case of strings maybe since it only relies on basic collection operations

However parameters such firstString/secondString can be a bit confusing so we could probably change it to something like firstSequence/secondSequence & finally add proper tests for numbers case

Do you think I am right here ?

[jordanmontt]

We have the same problem as described in here:
https://stackoverflow.com/questions/56597964/levenshtein-distance-between-list-of-number

So, if we want to apply it to numbers, which I believe that yes, we need to have special treatment for an array of numbers. I propose to create a class just for numbers and dispatch the operation to that class

[Alokzh]

@jordanmontt Sorry If I am not able to understand the problem you are trying to say, but I just went through the question of stackoverflow page & I think the issue there occurs due to conversion of array into strings & thus multi-digits such as #(14 20) gets treated as 4 different digits i.e (1, 4, 2, 0). So if we don't do any such conversion hopefully it will give correct result

Jordan I think if we create different class for array of numbers as you said the core algorithm might be nearly identical that's why I suggested just to change the name of the parameters but if you think we should have different class for array of numbers I can create a PR for that, please lmk

For example I tried to test the same example of stackoverflow & few more examples
Here we get correct result as expected & we don't get any errors

[jordanmontt]

Yes exactly. The problem is the conversion to string. For example converting to a string, the distance between 10 and 13 is 1. Which in reality needs to be 3

[Alokzh]

Yes exactly. The problem is the conversion to string. For example converting to a string, the distance between 10 and 13 is 1. Which in reality needs to be 3

But Jordan in this example you said i.e [10] & [13] why distance needs to be 3 , I mean don't you think according to levenshtein distance for this example we will get correct output whether we do conversion or don't do because in case of no conversion we can substitute 10 with 13 in one go & hence distance = 1 & if we do conversion to string even then we will get correct output i.e 1 because we will only need to substitute '0' with '3' after conversion

So Let's take a better example i.e [15] & [23] in this case we should expect the answer to be 1 according to levenshtein distance because we can substitute 15 with 23 while if we convert it to strings then the output we'll get is 2 because then we need to make 2 substitution i.e '1' with '2' & '5' with '3'

Do you think I am missing something or not able to understand ?

[Alokzh]

Also I just wanted to know that are you suggesting that we should have a class for array of numbers such that even if we do conversion to string we should get the correct output what we would get if we don't do any such conversion ?

But do you think is it necessary I mean if someone wants to find levenshtein distance between 2 arrays for numbers why should they convert to strings they can directly use this code & get the correct result as expected

distance := AILevenshteinDistance new.

Transcript show: 'Example: ' ;
	show: (distance distanceBetween: #(1 5 4 23) and: #(1 3 4 18)) asString; cr.

[jordanmontt]

So Let's take a better example i.e [15] & [23] in this case we should expect the answer to be 1 according to levenshtein distance because we can substitute 15 with 23 while if we convert it to strings then the output we'll get is 2 because then we need to make 2 substitution i.e '1' with '2' & '5' with '3'

for me this should be 8 if we are dealing with numbers, not 2.

[jordanmontt]

Also I just wanted to know that are you suggesting that we should have a class for array of numbers such that even if we do conversion to string we should get the correct output what we would get if we don't do any such conversion ?

no, if the users sends a string '15' that should be treated as a string.
'15' distanceBetween: '23' -> 2 Correct
but if they are NUMBERS
15 distanceBetween: 23 -> 8 Correct

[jordanmontt]

But we need to discuss this and to see that the other implementation in other languages do.

[Alokzh]

Alright Jordan now I get it, you are suggesting that in case of numbers it should return absolute difference