mln.py

from utility import read_mapping
from movielens.process import process_set
from sklearn import metrics
from math import exp
from statistics import mean, stdev

def mln(user, near_count, train_count, test_count, iterations):
    users = read_mapping("./process/users.txt")
    movies = read_mapping("./process/movies.txt")
    near = read_mapping("./process/near.txt")[user][:near_count]
    train = process_set(user=user, count=train_count, users=users, movies=movies, mechanism=None)
    test = process_set(user=user, count=test_count, users=users, movies=movies, mechanism=None)

    weights = [1 / near_count] * near_count
    em(weights, train, users, near, iterations)
    predictions = [predict(target[0], weights, users, near) for target in test]
    targets = [1 if target[1] else 0 for target in test]
    return metrics.roc_auc_score(targets, predictions)

def em(weights, train, users, near, iterations):
    for _ in range(iterations):
        predictions = [predict(target[0], weights, users, near) for target in train]
        weights = [reweigh(train, users[user], predictions) for user in near]
        total = sum(weights)
        weights = [weight / total for weight in weights]

def predict(movie, weights, users, near):
    score = 0
    for weight, user in zip(weights, near):
        score += weight if movie in users[user] else -weight
    return 1 / (1 + exp(-score))

def reweigh(train, user, predictions):
    weight = 0
    for target, prediction in zip(train, predictions):
        weight += prediction if target[0] in user else 1-prediction
    return weight

USER = "Ariel-Garcé"
PATH = f"./experiment/mln-{USER}"

a = [mln(user=USER, near_count=5, train_count=100, test_count=100, iterations=10) for _ in range(10)]
print(mean(a))
print(stdev(a))