Merge pull request #15 from LuminosoInsight/wordfreq-review

General style fixes and improvements from the code review

Former-commit-id: 8686a47a30
This commit is contained in:
Andrew Lin 2015-07-07 16:53:17 -04:00
commit ad165d2830
5 changed files with 155 additions and 149 deletions

92
scripts/gen_regex.py Normal file
View File

@ -0,0 +1,92 @@
import unicodedata
from ftfy import chardata
import pathlib
from pkg_resources import resource_filename
DATA_PATH = pathlib.Path(resource_filename('wordfreq', 'data'))
def cache_regex_from_func(filename, func):
"""
Generates a regex from a function that accepts a single unicode character,
and caches it in the data path at filename.
"""
with (DATA_PATH / filename).open(mode='w') as file:
file.write(func_to_regex(func))
def _emoji_char_class():
"""
Build a regex for emoji substitution. We create a regex character set
(like "[a-cv-z]") matching characters we consider emoji.
"""
cache_regex_from_func(
'emoji.txt',
lambda c:
chardata.CHAR_CLASS_STRING[ord(c)] == '3' and
c >= '\u2600' and c != '\ufffd'
)
def _non_punct_class():
"""
Builds a regex that matches anything that is not one of the following
classes:
- P: punctuation
- S: symbols
- Z: separators
- C: control characters
This will classify symbols, including emoji, as punctuation; callers that
want to treat emoji separately should filter them out first.
"""
cache_regex_from_func(
'non_punct.txt',
lambda c: unicodedata.category(c)[0] not in 'PSZC'
)
def _combining_mark_class():
"""
Builds a regex that matches anything that is a combining mark
"""
cache_regex_from_func(
'combining_mark.txt',
lambda c: unicodedata.category(c)[0] == 'M'
)
def func_to_regex(accept):
"""
Converts a function that accepts a single unicode character into a regex.
Unassigned unicode characters are treated like their neighbors.
"""
ranges = []
start = None
has_accepted = False
for x in range(0x110000):
c = chr(x)
if accept(c):
has_accepted = True
if start is None:
start = c
elif unicodedata.category(c) == 'Cn':
if start is None:
start = c
elif start is not None:
if has_accepted:
ranges.append('-'.join([start, chr(x-1)]))
has_accepted = False
start = None
else:
if has_accepted and start is not None:
ranges.append('-'.join([start, chr(x-1)]))
return '[%s]' % ''.join(ranges)
if __name__ == '__main__':
_combining_mark_class()
_non_punct_class()
_emoji_char_class()

View File

@ -1,9 +1,10 @@
from wordfreq import (
word_frequency, available_languages, cB_to_freq, iter_wordlist,
top_n_list, random_words, random_ascii_words, tokenize
word_frequency, available_languages, cB_to_freq,
top_n_list, random_words, random_ascii_words, tokenize,
half_harmonic_mean
)
from nose.tools import (
eq_, assert_almost_equal, assert_greater, assert_less, raises
eq_, assert_almost_equal, assert_greater, raises
)
@ -43,10 +44,10 @@ def test_twitter():
word_frequency('rt', lang, 'combined'))
def test_defaults():
def test_minimums():
eq_(word_frequency('esquivalience', 'en'), 0)
eq_(word_frequency('esquivalience', 'en', default=1e-6), 1e-6)
eq_(word_frequency('esquivalience', 'en', minimum=1e-6), 1e-6)
eq_(word_frequency('the', 'en', minimum=1), 1)
def test_most_common_words():
# If something causes the most common words in well-supported languages to
@ -96,7 +97,6 @@ def test_tokenization():
# We preserve apostrophes within words, so "can't" is a single word in the
# data, while the fake word "plan't" can't be found.
eq_(tokenize("can't", 'en'), ["can't"])
eq_(tokenize("plan't", 'en'), ["plan't"])
eq_(tokenize('😂test', 'en'), ['😂', 'test'])
@ -113,8 +113,13 @@ def test_casefolding():
def test_phrase_freq():
plant = word_frequency("plan.t", 'en')
assert_greater(plant, 0)
assert_less(plant, word_frequency('plan', 'en'))
assert_less(plant, word_frequency('t', 'en'))
assert_almost_equal(
plant,
half_harmonic_mean(
word_frequency('plan', 'en'),
word_frequency('t', 'en')
)
)
def test_not_really_random():
@ -132,3 +137,14 @@ def test_not_really_random():
@raises(ValueError)
def test_not_enough_ascii():
random_ascii_words(lang='zh')
def test_ar():
eq_(
tokenize('متــــــــعب', 'ar'),
['متعب']
)
eq_(
tokenize('حَرَكَات', 'ar'),
['حركات']
)

View File

@ -1,12 +1,10 @@
from pkg_resources import resource_filename
from functools import lru_cache
import unicodedata
from ftfy import chardata
import langcodes
import itertools
import msgpack
import re
import gzip
import itertools
import pathlib
import random
import logging
@ -16,125 +14,22 @@ DATA_PATH = pathlib.Path(resource_filename('wordfreq', 'data'))
CACHE_SIZE = 100000
def _emoji_char_class():
def load_range(filename):
"""
Build a regex for emoji substitution. First we create a regex character set
(like "[a-cv-z]") matching characters we consider emoji (see the docstring
of _replace_problem_text()). The final regex matches one such character
followed by any number of spaces and identical characters.
Loads a file from the data path
"""
ranges = []
for i, c in enumerate(chardata.CHAR_CLASS_STRING):
if c == '3' and i >= 0x2600 and i != 0xfffd:
if ranges and i == ranges[-1][1] + 1:
ranges[-1][1] = i
else:
ranges.append([i, i])
return '[%s]' % ''.join(chr(a) + '-' + chr(b) for a, b in ranges)
with (DATA_PATH / filename).open() as file:
return file.read()
EMOJI_RANGE = _emoji_char_class()
EMOJI_RANGE = load_range('emoji.txt')
NON_PUNCT_RANGE = load_range('non_punct.txt')
COMBINING_MARK_RANGE = load_range('combining_mark.txt')
def _non_punct_class():
"""
Builds a regex that matches anything that is not a one of the following
classes:
- P: punctuation
- S: symbols
- Z: separators
- C: control characters
This will classify symbols, including emoji, as punctuation; callers that
want to treat emoji separately should filter them out first.
"""
non_punct_file = DATA_PATH / 'non_punct.txt'
try:
with non_punct_file.open() as file:
return file.read()
except FileNotFoundError:
out = func_to_regex(lambda c: unicodedata.category(c)[0] not in 'PSZC')
with non_punct_file.open(mode='w') as file:
file.write(out)
return out
def _combining_mark_class():
"""
Builds a regex that matches anything that is a combining mark
"""
_combining_mark_file = DATA_PATH / 'combining_mark.txt'
try:
with _combining_mark_file.open() as file:
return file.read()
except FileNotFoundError:
out = func_to_regex(lambda c: unicodedata.category(c)[0] == 'M')
with _combining_mark_file.open(mode='w') as file:
file.write(out)
return out
def func_to_ranges(accept):
"""
Converts a function that accepts a single unicode character into a list of
ranges. Unassigned unicode are automatically accepted.
"""
ranges = []
start = None
for x in range(0x110000):
cat = unicodedata.category(chr(x))
if cat == 'Cn' or accept(chr(x)):
if start is None:
start = x
else:
if start is not None:
ranges.append((start, x-1))
start = None
if start is not None:
ranges.append((start, x))
return ranges
unassigned_ranges = None
def func_to_regex(accept):
"""
Converts a function that accepts a single unicode character into a regex.
Unassigned unicode characters are treated like their neighbors.
"""
ranges = []
start = None
for x in range(0x110000):
cat = unicodedata.category(chr(x))
if cat == 'Cn' or accept(chr(x)):
if start is None:
start = x
else:
if start is not None:
ranges.append((start, x-1))
start = None
if start is not None:
ranges.append((start, x))
global unassigned_ranges
if unassigned_ranges is None:
unassigned_ranges = set(func_to_ranges(lambda _: False))
ranges = [range for range in ranges if range not in unassigned_ranges]
return '[%s]' % ''.join("%s-%s" % (chr(start), chr(end))
for start, end in ranges)
COMBINING_MARK_RE = re.compile(_combining_mark_class())
NON_PUNCT_RANGE = _non_punct_class()
COMBINING_MARK_RE = re.compile(COMBINING_MARK_RANGE)
TOKEN_RE = re.compile("{0}|{1}+(?:'{1}+)*".format(EMOJI_RANGE, NON_PUNCT_RANGE))
def simple_tokenize(text):
"""
A simple tokenizer that can be applied to most languages.
@ -169,13 +64,11 @@ def tokenize(text, lang):
if mecab_tokenize is None:
from wordfreq.mecab import mecab_tokenize
return mecab_tokenize(text)
elif lang == 'ar':
tokens = simple_tokenize(text)
tokens = [token.replace('ـ', '') for token in tokens] # remove tatweel
tokens = [COMBINING_MARK_RE.sub('', token) for token in tokens]
return [token for token in tokens if token] # remove empty strings
else:
return simple_tokenize(text)
if lang == 'ar':
text = COMBINING_MARK_RE.sub('', text.replace('ـ', ''))
return simple_tokenize(text)
def read_cBpack(filename):
@ -284,7 +177,7 @@ def cB_to_freq(cB):
"""
if cB > 0:
raise ValueError(
"A frequency cannot be a positive number of decibels."
"A frequency cannot be a positive number of centibels."
)
return 10 ** (cB / 100)
@ -298,8 +191,9 @@ def get_frequency_dict(lang, wordlist='combined', match_cutoff=30):
freqs = {}
pack = get_frequency_list(lang, wordlist, match_cutoff)
for index, bucket in enumerate(pack):
freq = cB_to_freq(-index)
for word in bucket:
freqs[word] = cB_to_freq(-index)
freqs[word] = freq
return freqs
@ -312,8 +206,7 @@ def iter_wordlist(lang, wordlist='combined'):
with the same rounded frequency, appearing in alphabetical order within
each band.
"""
for sublist in get_frequency_list(lang, wordlist):
yield from sublist
return itertools.chain(*get_frequency_list(lang, wordlist))
def half_harmonic_mean(a, b):
@ -328,25 +221,26 @@ def half_harmonic_mean(a, b):
@lru_cache(maxsize=CACHE_SIZE)
def word_frequency(word, lang, wordlist='combined', default=0.):
def word_frequency(word, lang, wordlist='combined', minimum=0.):
"""
Get the frequency of `word` in the language with code `lang`, from the
specified `wordlist`. The default wordlist is 'combined', built from
whichever of these four sources have sufficient data for the language:
whichever of these five sources have sufficient data for the language:
- Full text of Wikipedia
- A sample of 72 million tweets collected from Twitter in 2014,
divided roughly into languages using automatic language detection
- Frequencies extracted from OpenSubtitles
- The Leeds Internet Corpus
- Google Books Ngrams and Google Books Syntactic Ngrams
Another available wordlist is 'twitter', which uses only the data from
Twitter.
Words that we believe occur at least once per million tokens, based on
the average of these lists, will appear in the word frequency list.
If you look up a word that's not in the list, you'll get the `default`
value, which itself defaults to 0.
The value returned will always be at least as large as `minimum`.
If a word decomposes into multiple tokens, we'll return a smoothed estimate
of the word frequency that is no greater than the frequency of any of its
@ -357,12 +251,12 @@ def word_frequency(word, lang, wordlist='combined', default=0.):
tokens = tokenize(word, lang)
if len(tokens) == 0:
return default
return minimum
for token in tokens:
if token not in freqs:
# If any word is missing, just return the default value
return default
return minimum
value = freqs[token]
if combined_value is None:
combined_value = value
@ -370,11 +264,16 @@ def word_frequency(word, lang, wordlist='combined', default=0.):
# Combine word values using the half-harmonic-mean formula,
# (a * b) / (a + b). This operation is associative.
combined_value = half_harmonic_mean(combined_value, value)
return combined_value
return max(combined_value, minimum)
@lru_cache(maxsize=100)
def top_n_list(lang, n, wordlist='combined', ascii_only=False):
"""
Return a frequency list of length `n` in descending order of frequency.
This list contains words from `wordlist`, of the given language.
If `ascii_only`, then only ascii words are considered.
"""
results = []
for word in iter_wordlist(lang, wordlist):
if (not ascii_only) or max(word) <= '~':
@ -384,7 +283,7 @@ def top_n_list(lang, n, wordlist='combined', ascii_only=False):
return results
def random_words(lang='en', wordlist='combined', nwords=4, bits_per_word=12,
def random_words(lang='en', wordlist='combined', nwords=5, bits_per_word=12,
ascii_only=False):
"""
Returns a string of random, space separated words.
@ -410,7 +309,7 @@ def random_words(lang='en', wordlist='combined', nwords=4, bits_per_word=12,
return ' '.join(selected)
def random_ascii_words(lang='en', wordlist='combined', nwords=4,
def random_ascii_words(lang='en', wordlist='combined', nwords=5,
bits_per_word=12):
"""
Returns a string of random, space separated, ASCII words.

1
wordfreq/data/emoji.txt Normal file
View File

@ -0,0 +1 @@
[☀-♮♰-❧➔-➿⠀-⣿⬀-⬯⭅-⭆⭍-⯿⳥-⳪⸼-⿿〄-〄〒-〓〠-〠〶-〷〾-぀㆏-㆑㆖-㆟ㆻ-㇯㈀-㈟㈪-㉇㉐-㉐㉠-㉿㊊-㊰㋀-㏿䶶-䷿꒍-꓏꠨-꠯꠶-꠷꠹-꠿꩷-꩹﷽-﷿¦-¦￧-│■--𐄴-𐄿𐅹-𐆉𐆋-𐇼𐡠-𐣿𐪀-𐫿𖨹-𖻿𛀂-𝅘𝅥𝅲𝅪-𝅬𝆃-𝆄𝆌-𝆩𝆮-𝉁𝉅-𝍟𞻲-🃿🄋-🿿]

View File

@ -14,8 +14,6 @@ def mecab_tokenize(text):
contains the same table that the command-line version of MeCab would output.
We find the tokens in the first column of this table.
"""
parsed_str = MECAB_ANALYZER.parse(text.strip())
lines = [line for line in parsed_str.split('\n')
if line != '' and line != 'EOS']
tokens = [line.split('\t')[0] for line in lines]
return tokens
return [line.split('\t')[0]
for line in MECAB_ANALYZER.parse(text.strip()).split('\n')
if line != '' and line != 'EOS']