mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-27 15:01:08 +00:00
61 lines
1.4 KiB
Python
61 lines
1.4 KiB
Python
import os
|
|
|
|
UPPERLETTERS = "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
|
|
LETTERS_AND_SPACE = UPPERLETTERS + UPPERLETTERS.lower() + " \t\n"
|
|
|
|
|
|
def loadDictionary():
|
|
path = os.path.split(os.path.realpath(__file__))
|
|
englishWords = {}
|
|
with open(path[0] + "/dictionary.txt") as dictionaryFile:
|
|
for word in dictionaryFile.read().split("\n"):
|
|
englishWords[word] = None
|
|
return englishWords
|
|
|
|
|
|
ENGLISH_WORDS = loadDictionary()
|
|
|
|
|
|
def getEnglishCount(message):
|
|
message = message.upper()
|
|
message = removeNonLetters(message)
|
|
possibleWords = message.split()
|
|
|
|
if possibleWords == []:
|
|
return 0.0
|
|
|
|
matches = 0
|
|
for word in possibleWords:
|
|
if word in ENGLISH_WORDS:
|
|
matches += 1
|
|
|
|
return float(matches) / len(possibleWords)
|
|
|
|
|
|
def removeNonLetters(message):
|
|
lettersOnly = []
|
|
for symbol in message:
|
|
if symbol in LETTERS_AND_SPACE:
|
|
lettersOnly.append(symbol)
|
|
return "".join(lettersOnly)
|
|
|
|
|
|
def isEnglish(message, wordPercentage=20, letterPercentage=85):
|
|
"""
|
|
>>> isEnglish('Hello World')
|
|
True
|
|
|
|
>>> isEnglish('llold HorWd')
|
|
False
|
|
"""
|
|
wordsMatch = getEnglishCount(message) * 100 >= wordPercentage
|
|
numLetters = len(removeNonLetters(message))
|
|
messageLettersPercentage = (float(numLetters) / len(message)) * 100
|
|
lettersMatch = messageLettersPercentage >= letterPercentage
|
|
return wordsMatch and lettersMatch
|
|
|
|
|
|
import doctest
|
|
|
|
doctest.testmod()
|