maxbachmann
commented
10 months ago Why do you consider the score 46 as incorrect? I understand you would like to find Real Madrid as best match for real, but according to the normalized Indel similarity it simply isn't.
To convert from real to Barcelona the following operations are required:
__r_e___al
Barcelona_These are 7 operation. Now this is normalized in the following way: 1.0 - (dist / (len(s1) + len(s2))) -> 0.46.
On the other hand for Real Madrid the following operations are required:
_real_______
R_eal MadridThese are 9 operations. Normalized in the same way you get 0.4.
Note, without python-Levenshtein install at all (clean, checked with colab with and without) it works like a charm.
No clue how that would work, since in this specific case the difflib fallback actually returns the same results:
>>> import difflib
>>> difflib.SequenceMatcher(None, 'real', 'Barcelona').ratio()
0.46153846153846156
>>> difflib.SequenceMatcher(None, 'real', 'Real Madrid').ratio()
0.4
orellavie1212
Hey, After testing failure of fuzz.ratio at the detection of basic sim. match, I found that the newer ver of python-Levenshtein(0.22.0) makes the fuzz.ratio output wrong answer when installing it with fuzzywuzzy (latest version of fuzzy). Here is an example of the wrong output: code: from fuzzywuzzy import fuzz list_to_search_from = ['Barcelona', 'Real Madrid'] text_to_search = 'real' scores = {search_option: fuzz.ratio(search_option, text_to_search) for search_option in list_to_search_from} output: {‘Barcelona’: 46, ‘Real Madrid’: 40} with ver 21.1 {'Barcelona': 31, 'Real Madrid': 40} which is the right answer.
After downgrade to 21.1 it works nice, didn't find where it fails on 22. Note, without python-Levenshtein install at all (clean, checked with colab with and without) it works like a charm. Wonder why it happens (maybe wrong calc of Lev. dist at the new ver) !pip install python-Levenshtein==0.21.1