mass-scraper/extractor/find_hours.py

57 lines
1.8 KiB
Python
Raw Normal View History

import re
from colorama import Fore, Back, Style
hour_regex = re.compile(
'(0?[6-9]|1\d|2[0-2])[:.](oo|[0-5]\d)|6|7|8|9|1\d|2[0-2]')
def borders_ok(text, start, end):
text = ' ' + text + ' '
before_start_char = text[start]
after_end_char = text[end + 1]
if ((before_start_char.isspace() or before_start_char in ',(/')
and (after_end_char.isspace() or after_end_char in ',;)/')
and (before_start_char != '(' or after_end_char != ')')):
return True
else:
return False
def delete_duplicates(text):
text = re.sub(' +', ' ', text)
text = re.sub(' ?\n ?', '\n', text)
text = re.sub('\n{5,}', '\n\n\n', text)
text = re.sub('\n\n', '\n', text)
return text
def get_context(text, start, end, minsize):
hour = text[start:end]
prefix = delete_duplicates(text[:start]).rsplit(
' ', maxsplit=minsize + 12)[1:]
suffix = delete_duplicates(text[end:]).split(
' ', maxsplit=minsize + 2)[:-1]
return ' '.join(prefix), hour, ' '.join(suffix)
def hours_iterator(text, minsize=20, color=False):
for hour_match in hour_regex.finditer(text):
start = hour_match.start(0)
end = hour_match.end(0)
if not borders_ok(text, start, end):
continue
prefix, hour, suffix = get_context(text, start, end, minsize)
if color:
utterance = f'{prefix}&&&{hour}###{suffix}'
yield utterance, color_hour(prefix, hour, suffix, Fore.GREEN,
Style.BRIGHT)
else:
yield prefix, hour, suffix
# w klasyfikatorze dzielić tak aby jeszcze \n było oddzielnie
def color_hour(prefix, hour, suffix, color=Fore.GREEN, style=Style.BRIGHT):
return prefix + color + style + hour + Style.RESET_ALL + suffix