Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

problem of fuzz.ratio with newer ver (22) of python-Lev. distance #334

Open
orellavie1212 opened this issue Oct 1, 2023 · 2 comments
Open

Comments

@orellavie1212
Copy link

orellavie1212 commented Oct 1, 2023

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

@maxbachmann
Copy link

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 Madrid

These 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
Copy link
Author

orellavie1212 commented Oct 1, 2023

after investigation you are indeed right, how do you offer to resolve such problem?
partial_ratio is not an offer as "Real Madrid" and "Real Saragosa" are the same for "Real", and their partial_ratio rate should be the same.
I wonder how to attack such problem

I went with fuzz.partial_token_set_ratio, as it should fit better I think

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants