bazarr/libs/textdistance/algorithms/phonetic.py

180 lines
6.0 KiB
Python

from __future__ import annotations
# built-in
from collections import defaultdict
from itertools import groupby, zip_longest
from typing import Any, Iterator, Sequence, TypeVar
# app
from .base import Base as _Base, BaseSimilarity as _BaseSimilarity
try:
# external
import numpy
except ImportError:
numpy = None # type: ignore[assignment]
__all__ = [
'MRA', 'Editex',
'mra', 'editex',
]
T = TypeVar('T')
class MRA(_BaseSimilarity):
"""Western Airlines Surname Match Rating Algorithm comparison rating
https://en.wikipedia.org/wiki/Match_rating_approach
https://github.com/Yomguithereal/talisman/blob/master/src/metrics/mra.js
"""
def maximum(self, *sequences: str) -> int:
sequences = [list(self._calc_mra(s)) for s in sequences]
return max(map(len, sequences))
def _calc_mra(self, word: str) -> str:
if not word:
return word
word = word.upper()
word = word[0] + ''.join(c for c in word[1:] if c not in 'AEIOU')
# remove repeats like an UNIX uniq
word = ''.join(char for char, _ in groupby(word))
if len(word) > 6:
return word[:3] + word[-3:]
return word
def __call__(self, *sequences: str) -> int:
if not all(sequences):
return 0
sequences = [list(self._calc_mra(s)) for s in sequences]
lengths = list(map(len, sequences))
count = len(lengths)
max_length = max(lengths)
if abs(max_length - min(lengths)) > count:
return 0
for _ in range(count):
new_sequences = []
minlen = min(lengths)
for chars in zip(*sequences):
if not self._ident(*chars):
new_sequences.append(chars)
new_sequences = map(list, zip(*new_sequences))
# update sequences
ss: Iterator[tuple[Any, Any]]
ss = zip_longest(new_sequences, sequences, fillvalue=list())
sequences = [s1 + s2[minlen:] for s1, s2 in ss]
# update lengths
lengths = list(map(len, sequences))
if not lengths:
return max_length
return max_length - max(lengths)
class Editex(_Base):
"""
https://anhaidgroup.github.io/py_stringmatching/v0.3.x/Editex.html
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.14.3856&rep=rep1&type=pdf
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.18.2138&rep=rep1&type=pdf
https://github.com/chrislit/blob/master/abydos/distance/_editex.py
https://habr.com/ru/post/331174/ (RUS)
"""
groups: tuple[frozenset[str], ...] = (
frozenset('AEIOUY'),
frozenset('BP'),
frozenset('CKQ'),
frozenset('DT'),
frozenset('LR'),
frozenset('MN'),
frozenset('GJ'),
frozenset('FPV'),
frozenset('SXZ'),
frozenset('CSZ'),
)
ungrouped = frozenset('HW') # all letters in alphabet that not presented in `grouped`
def __init__(
self,
local: bool = False,
match_cost: int = 0,
group_cost: int = 1,
mismatch_cost: int = 2,
groups: tuple[frozenset[str], ...] = None,
ungrouped: frozenset[str] = None,
external: bool = True,
) -> None:
# Ensure that match_cost <= group_cost <= mismatch_cost
self.match_cost = match_cost
self.group_cost = max(group_cost, self.match_cost)
self.mismatch_cost = max(mismatch_cost, self.group_cost)
self.local = local
self.external = external
if groups is not None:
if ungrouped is None:
raise ValueError('`ungrouped` argument required with `groups`')
self.groups = groups
self.ungrouped = ungrouped
self.grouped = frozenset.union(*self.groups)
def maximum(self, *sequences: Sequence) -> int:
return max(map(len, sequences)) * self.mismatch_cost
def r_cost(self, *elements: str) -> int:
if self._ident(*elements):
return self.match_cost
if any(map(lambda x: x not in self.grouped, elements)):
return self.mismatch_cost
for group in self.groups:
if all(map(lambda x: x in group, elements)):
return self.group_cost
return self.mismatch_cost
def d_cost(self, *elements: str) -> int:
if not self._ident(*elements) and elements[0] in self.ungrouped:
return self.group_cost
return self.r_cost(*elements)
def __call__(self, s1: str, s2: str) -> float:
result = self.quick_answer(s1, s2)
if result is not None:
return result
# must do `upper` before getting length because some one-char lowercase glyphs
# are represented as two chars in uppercase.
# This might result in a distance that is greater than the maximum
# input sequence length, though, so we save that maximum first.
max_length = self.maximum(s1, s2)
s1 = ' ' + s1.upper()
s2 = ' ' + s2.upper()
len_s1 = len(s1) - 1
len_s2 = len(s2) - 1
d_mat: Any
if numpy:
d_mat = numpy.zeros((len_s1 + 1, len_s2 + 1), dtype=int)
else:
d_mat = defaultdict(lambda: defaultdict(int))
if not self.local:
for i in range(1, len_s1 + 1):
d_mat[i][0] = d_mat[i - 1][0] + self.d_cost(s1[i - 1], s1[i])
for j in range(1, len_s2 + 1):
d_mat[0][j] = d_mat[0][j - 1] + self.d_cost(s2[j - 1], s2[j])
for i, (cs1_prev, cs1_curr) in enumerate(zip(s1, s1[1:]), start=1):
for j, (cs2_prev, cs2_curr) in enumerate(zip(s2, s2[1:]), start=1):
d_mat[i][j] = min(
d_mat[i - 1][j] + self.d_cost(cs1_prev, cs1_curr),
d_mat[i][j - 1] + self.d_cost(cs2_prev, cs2_curr),
d_mat[i - 1][j - 1] + self.r_cost(cs1_curr, cs2_curr),
)
distance = d_mat[len_s1][len_s2]
return min(distance, max_length)
mra = MRA()
editex = Editex()