partial_ratio.py

from textdistance import damerau_levenshtein
import mojimoji
import regex as re

def partial_ratio(s1, s2, levenshtein_ratio=damerau_levenshtein.normalized_similarity):
    def _preprocess(s):
        s = mojimoji.zen_to_han(s, kana=False, ascii=True, digit=True)
        s = s.lower()
        s = re.sub('\s+', '', s)
        return s
    def _match_substring(shorter, longer):
        window = len(shorter)
        rng = len(longer) - window
        if rng>0:
            scores = [levenshtein_ratio(shorter, longer[i:i+window]) for i in range(rng)]
            # max_ix = np.argmax(scores)
            return max(scores)
        else:
            return 0.
    s1 = _preprocess(s1)
    s2 = _preprocess(s2)
    shorter, longer = s1, s2
    if len(s1) == len(s2):
        return levenshtein_ratio(s1, s2)
    elif len(s1) > len(s2):
        shorter, longer = s2, s1
    max_score = _match_substring(shorter, longer)
    return max_score

def fuzzy_difference(gs, ps, sim=lambda x,y:int(x==y), threshold=1.):
    # gs - ps
    diffs = set()
    for g in gs:
        if max([sim(g, p) for p in ps]) < threshold:  # g not in p
            diffs.add(g)
    return diffs