def cleanup_title(value):
# Need to use this rather than .title() because .title()
# does not handle things like "Wouldn't" properly. It
# converts it to "Wouldn'T" rather than keeping the T
# lowercase
if value[0] == '"' or value[0] == "'":
value = value[1:]
if value[len(value) - 1] == '"' or value[len(value) - 1] == "'":
value = value[:len(value) - 1]
value = value.replace('"', "").strip()
value = HTMLParser.HTMLParser().unescape(value.lower())
en_us_locale = icu.Locale('en_US')
break_iter = icu.BreakIterator.createTitleInstance(en_us_locale)
temp_title = icu.UnicodeString(value)
title = unicode(temp_title.toTitle(break_iter, en_us_locale))
word_enders = [" ", ",", ".", ";", ":", '"', "'", "-"]
for acronym in settings.COMPANY_ACRONYMS:
if '.com' in acronym[0]:
# .com often comes at the end of a title so we don't want to add
# the trailing space check
if acronym[1] in title:
title = title.replace(acronym[1], acronym[0])
else:
if title.rfind(acronym[1]) == len(title) - len(acronym[1]):
title = "%s%s" % (title[:len(title) -
(len(acronym[1]))], acronym[0])
for ender in word_enders:
if "%s%s" % (acronym[1], ender) in title:
title = title.replace("%s%s" % (acronym[1], ender),
"%s%s" % (acronym[0], ender))
return title
def _normalize(bs: bistr, normalizer: icu.Normalizer2) -> bistr:
builder = BistrBuilder(bs)
us = icu.UnicodeString(bs.modified)
offset = 0
while not builder.is_complete:
i = normalizer.spanQuickCheckYes(us)
builder.skip(us.countChar32(0, i))
if builder.is_complete:
break
us = us[i:]
i = 0
while i < len(us):
if us.charAt(i) & 0xFC00 == 0xD800:
i += 1
i += 1
if normalizer.hasBoundaryBefore(chr(us.char32At(i))):
break
chunk = us[:i]
normalized = str(normalizer.normalize(chunk))
builder.replace(chunk.countChar32(), normalized)
us = us[i:]
return builder.build()
def __init__(self):
# Graph where the nodes are unicode characters and the edges are "contains"
# such that successors(尔) = [...你...]., and predecessors(你) = [亻,尔].
# So, insert with self._graph.add_edge( "亻", "你" )
# self._graph.add_edge( "尔", "你" )
self._graph = nx.DiGraph()
with open(babelstone.PATH_TO_IDS_TXT, encoding="UTF-8") as fp:
for line in fp:
# Ignore comments
if line.startswith("#"):
continue
# TODO(ambuc): ids.txt uses:
# {1}, {2}, etc. to represent unencoded components.
# ↔ as a mirror operator, i.e. to represent a component without
# a Unicode encoding, but whose mirror does have a Unicode
# encoding.
# ↷ as a rotation operator, i.e. to represent a component
# without a Unicode encoding, but whose 180deg rotation does
# have a Unicode encoding.
# 〾 as a variation indicator. We should try to handle these.
# ?, ？ ids.txt uses these to represent an unencodable component.
# We should probably try to handle these edge cases.
elif re.search("[{}↔↷〾?？]", line):
continue
maybe_parsed_set = parse(str(icu.UnicodeString(line)))
if maybe_parsed_set is not None:
self.insert(maybe_parsed_set)
def get_section_title(ch):
nkfd_form = unicodedata.normalize('NFKD', unicode(ch))
nkfd_ch = nkfd_form[0]
cat = unicodedata.category(nkfd_ch)
if 'L' != cat[0]: # Not a letter
return ''
if 'l' != cat[1]: # Not a lower-case letter (uppercase or special)
return nkfd_ch
return unicode(icu.UnicodeString(nkfd_ch).toUpper(lang_locale))
def _edit(bs: bistr, op: Callable, locale: Optional[str] = None) -> bistr:
builder = BistrBuilder(bs)
edits = icu.Edits()
ucur = icu.UnicodeString(builder.current)
if locale is None:
umod = icu.UnicodeString(op(ucur, edits))
else:
umod = icu.UnicodeString(op(icu.Locale(locale), ucur, edits))
for is_change, old_len, new_len, old_i, new_i, _ in edits.getFineIterator():
old_len = ucur.countChar32(old_i, old_len)
if is_change:
repl = str(umod[new_i:new_i+new_len])
builder.replace(old_len, repl)
else:
builder.skip(old_len)
return builder.build()
def tokenize(self, text: String) -> Tokenization:
text = bistr(text)
tokens = []
bi = self._break_iterator()
utext = icu.UnicodeString(text.modified)
bi.setText(utext)
ui = bi.first()
uj = bi.nextBoundary()
i = 0
while uj != icu.BreakIterator.DONE:
j = i + utext.countChar32(ui, uj - ui)
if self._check_token(bi.getRuleStatus()):
tokens.append(Token.slice(text, i, j))
ui = uj
uj = bi.nextBoundary()
i = j
return Tokenization(text, tokens)
regex_doublequotes = re.compile(r'\"+');
"""
To detect Multiword phrases inside a sentence we are using Gensim Library and it's models.phrases class
https://radimrehurek.com/gensim/models/phrases.html. We have already made an object of the class and put all the tokenized reviews
in it and stored it in gensim_mulitphrases class (see generate_pickle_for_Gensim.py). Now we are loading object back from
pickle file
"""
with open("gensim_mulitphrases_"+Business+".txt", "rb") as fp:
bigram = pickle.load(fp)
"""
The py2casefold in not tested enough, slow, and I couldn't install it in anaconda. This lambda function is a replacement.
The function returns a string, not a unicode, because we will use the .translate fast function to remove punct. of strings.
Source: https://stackoverflow.com/a/32838944/3429115
"""
casefold = lambda u: unicode(icu.UnicodeString(u).foldCase()).encode(CharsSet,"replace");
CharsSet = "ascii"; # The Character set to be used as the default one when interpreting texts
def getWordnetPos(_treebank_tag):
"""
Translate the tree bank PoS tags to the WordNet's
> Parameters:
_treebank_tag : str | The tag to be translated
> Returns:
The relevant WordNet PoS tag
https://stackoverflow.com/a/15590384/3429115
"""
if _treebank_tag.startswith('J'):
return wordnet.ADJ
spec = "".join(file(opts.input).readlines())
brk = icu.RuleBasedBreakIterator(spec)
else:
brk = icu.RuleBasedBreakIterator()
print(brk.getRules())
if opts.codes:
text = "".join(chr(int(x, 16)) for x in args)
elif opts.file:
text = "".join(file(args[0]).readlines())
else:
text = args[0]
res = []
brk.setText(icu.UnicodeString(text))
last = brk.first()
try:
while True:
next = brk.next()
# print(next, " ", brk.getRuleStatus())
res.append(text[last:next])
last = next
except:
res.append(text[last:])
if opts.hex:
print(f" {opts.separator} ".join(" ".join(hex(ord(x)) for x in res)
for y in res))
else:
print(opts.separator.join(res))
import random
# The path to the main output folder, in which the text files are placed, and checks are made to avoid duplicated work:
directory = "./RestaurantsToOthers_Data"
regex_slash_newlines = re.compile(r'[\n\r\\]+')
regex_slash_tabs = re.compile(r'\t+')
regex_doublequotes = re.compile(r'\"+')
"""
The py2casefold in not tested enough, slow, and I couldn't install it in anaconda. This lambda function is a replacement.
The function returns a string, not a unicode, because we will use the .translate fast function to remove punct. of strings.
Source: https://stackoverflow.com/a/32838944/3429115
"""
CharsSet = "ascii"
# The Character set to be used as the default one when interpreting texts
casefold = lambda u: unicode(icu.UnicodeString(u).foldCase()).encode(
CharsSet, "ignore")
def iter_sample_fast(iterable, samplesize):
"""
Fast memory-efficient sampling method for pretty large iterables.
Adopted from: https://stackoverflow.com/a/12583436/3429115
> Parameters:
* iterable: iterable object | The collection we want to sample from
* samplesize: int | How many samples to generate
> Returns:
List of samples, hoewver, since sampling is made without replacement (most probably), they aren't IID; but