To make all of our hard work easier to use, we need to pack it all up into a single, neat function, as shown:

# remake a simple two char CV
two_cv = CountVectorizer(ngram_range=(1, 2), analyzer='char', lowercase=False)

two_char = two_cv.fit_transform(text)

two_char
# there are 7,528 unique 2-in-a-row-chars (number of columns)

<1048485x7528 sparse matrix of type '<type 'numpy.int64'>'
  with 14350326 stored elements in Compressed Sparse Row format>


# make a simple function using the two_char CV and matrix
def get_closest_word_similarity(password):
 raw_vectorization = cosine_similarity(two_cv.transform([password]), two_char)
 return raw_vectorization[:,np.argsort(raw_vectorization)[0,-20:]].mean()

This function makes it easier to judge passwords quickly:

print get_closest_word_similarity("guest123") # very close to passwords in the db

0.789113817

print get_closest_word_similarity("sdfFSKSJNDFKFSD3253245sadSDF@@$@#$") # not very close to passwords in the db

0.47148393

We can take this one step further and create a custom password-tester class that will store in-memory vectorizations of passwords to make our algorithm easy to share:

# this is a complete data-driven automated password strength tester that judges passwords without any human intuition.

class PasswordTester():
    def __init__(self, text):
        self.vectorizer = None
        self.password_matrix = None
        self.text = text

    def make_vectorizer(self, **kwargs):
        self.vectorizer = CountVectorizer(**kwargs)
        self.password_matrix = self.vectorizer.fit_transform(self.text)

    def get_closest_word_similarity(self, password):
        raw_vectorization = cosine_similarity(self.vectorizer.transform([password]), self.password_matrix)
        return raw_vectorization[:,np.argsort(raw_vectorization)[0,-20:]].mean()

    def judge_password(self, attempted_password):
        badness_score = self.get_closest_word_similarity(attempted_password)
        if badness_score > .9:
            return "very poor", badness_score
        elif badness_score > .8:
            return "poor", badness_score
        elif badness_score > .6:
            return "not bad", badness_score
        elif badness_score > .4:
            return "good", badness_score
        else:
            return "very good", badness_score

To use our custom class, we can instantiate it with custom vectorization parameters:

p = PasswordTester(text)
p.make_vectorizer(ngram_range=(1, 2), analyzer='char', lowercase=False)


p.judge_password("password123321")
('poor', 0.8624222257655552)

p.judge_password("Istanbul9999")
('not bad', 0.7928432151071905)

# generated from LastPass, a password management and creation service 10 digit
p.judge_password("D9GLRyG0*!")
('good', 0.41329460236856164)

# generated from LastPass, 100 digit
p.judge_password("ES%9G1UxtoBlwn^e&Bz3bAj2hMfk!2cfj8kF8yUc&J2B&khzNpBoe65Va!*XGXH1&PF5fxbKGpBsvPNQdnmnWyzb@W$tcn^%fnKa")
('very good', 0.3628996523892102)