Создание системы рекомендаций from collections import Counter import datetime import numpy as np from sklearn.ensemble import RandomForestClassifier from sklearn.utils import shuffle from models.features.topics import topics_similarity from models.features.cosine import cosine_similarity_features clf = RandomForestClassifier() clf_one = RandomForestClassifier() clf_two = RandomForestClassifier() def generate_features(data, val=None): features = [] for raw in data: features.extend(topics_similarity(raw)) features.extend(cosine_similarity_features(data[:-1], data[-1])) if val is None: return features else: return features, val def generate_data(data): x_true = [] y_true = [] x_false = [] y_false = [] print('Start generate features.') for urls in data.sequences.find(): features = generate_features(data.get_articles_data(urls['urls']), 1) x_true.append(features[0]) y_true.append(features[1]) print('Start generate random data.') while len(x_true) != len(x_false): features = generate_features(data.get_random_articles(4), 0) x_false.append(features[0]) y_false.append(features[1]) return x_true + x_false, y_true + y_false def init(data): try: x = np.load(open('train_x.np', 'rb')) y = np.load(open('train_y.np', 'rb')) except FileNotFoundError: x, y = generate_data(data) np.save(open('train_x.np', 'wb'), x) np.save(open('train_y.np', 'wb'), y) x, y = shuffle(x, y) x_one = list(map(lambda a: a[:81], x)) x_two = list(map(lambda a: a[:122], x)) print('Train model for 3 articles.') clf.fit(x, y) print('Train model for 1 article.') clf_one.fit(x_one, y) print('Train model for 2 articles.') clf_two.fit(x_two, y) def predict(input_articles, input_ids,tvrain_data, recommends_num): result_counter = Counter() min_time = input_articles[0]['time'] - datetime.timedelta(hours=5) max_time = input_articles[-1]['time'] + datetime.timedelta(hours=5) mongo_query = {'time': {'$gt': min_time, '$lt': max_time}} len_articles = len(input_articles) # Gen for articles in Mongo for article in tvrain_data.iterate_articles(except_articles=input_ids, query=mongo_query): new_features = generate_features(input_articles + [article]) if len_articles == 3: result = clf.predict_proba([new_features]) elif len_articles == 2: result = clf_two.predict_proba([new_features]) elif len_articles == 1: result = clf_one.predict_proba([new_features]) result_counter[article['_id']] = result[0][1] return list(result_counter.most_common(recommends_num)) Материалы по данной теме