fuzz.ratio using levenstein distance?

[shubhamscifi]

How to calculate fuzz.ratio using levenstein distance? The current implementation uses only Indel distance.

[maxbachmann]

You can use the Levenshtein module:

from rapidfuzz.distance import Levenshtein

score = Levenshtein.normalized_similarity(s1, s2)

which returns the normalized similarity as score between 0 and 1.

[shubhamscifi]

I want to use rapidfuzz.process.extractOne also using Levenshtein distance. Which I am guessing, uses vectorized operations and hence will be faster. How can I do this?

[maxbachmann]

You can use it the same way you would with fuzz.ratio:

from rapidfuzz.distance import Levenshtein
from rapidfuzz import process

process.extractOne(query, choices, scorer=Levenshtein.normalized_similarity)

[shubhamscifi]

The output of thefuzz levenshtein and rapidfuzz doesn't match even after rounding off. Why?

a = 'abcd'
b = 'abcde'
Indel.normalized_similarity(a,b)*100    # 88.88888888888889
Levenshtein.normalized_similarity(a,b)*100  # 80.0
thefuzz.fuzz.ratio(a,b)     # 89

[maxbachmann]

fuzzywuzzy/thefuzz uses the Levenshtein distance with the following weights: Insertion=1, Deletion=1, Subsitution=2. This is equivalent to the Indel distance, since any substitution can be represented by an Insertion + Deletion. You can achieve the same results using the weighted Levenshtein distance in RapidFuzz:

a = 'abcd'
b = 'abcde'
Levenshtein.normalized_similarity(a,b, weights=(1,1,2))*100  # 88.88888888888889

[shubhamscifi]

Thank you!