scoring.py

from sklearn.feature_extraction.text import TfidfVectorizer
import numpy as np

def calc_similarity_scores(strings):
    """Returns an l2 normalized vector of similarity scores corresponding to the
    strings.

    :param list strings: The collection of strings for which the function
    computes similarity scores.
    """
    # Find Tfidf pairwise scores
    vect = TfidfVectorizer(min_df=1)
    tfidf = vect.fit_transform(strings)
    pairwise_scores = (tfidf * tfidf.T).A
    # Convert pairwise scores into 1 score for each string
    # Normalize scores
    scores = np.array([np.linalg.norm(vec) for vec in pairwise_scores])
    return scores / np.linalg.norm(scores)

def calc_popularity(scores):
    """Returns an array of integer popularities [1, 10] based on percentile
    w.r.t other scores.

    :param list float: The collection of scores for which the function computes
    popularity
    """
    quantiles = [10*i for i in range(1,10)]
    percentiles = np.percentile(scores, quantiles)
    return np.digitize(scores, percentiles) + 1