/
svd.py
65 lines (54 loc) · 2.97 KB
/
svd.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
from util.models import RecommendResult, Dataset
from src.base_recommender import BaseRecommender
from collections import defaultdict
import scipy
import numpy as np
np.random.seed(0)
class SVDRecommender(BaseRecommender):
def recommend(self, dataset: Dataset, **kwargs) -> RecommendResult:
# 欠損値の穴埋め方法
fillna_with_zero = kwargs.get("fillna_with_zero", True)
factors = kwargs.get("factors", 5)
# 評価値をユーザー×映画の行列に変換。欠損値は、平均値または0で穴埋めする
user_movie_matrix = dataset.train.pivot(index="user_id", columns="movie_id", values="rating")
user_id2index = dict(zip(user_movie_matrix.index, range(len(user_movie_matrix.index))))
movie_id2index = dict(zip(user_movie_matrix.columns, range(len(user_movie_matrix.columns))))
if fillna_with_zero:
matrix = user_movie_matrix.fillna(0).to_numpy()
else:
matrix = user_movie_matrix.fillna(dataset.train.rating.mean()).to_numpy()
# 因子数kで特異値分解を行う
P, S, Qt = scipy.sparse.linalg.svds(matrix, k=factors)
# 予測評価値行列
pred_matrix = np.dot(np.dot(P, np.diag(S)), Qt)
# 学習用に出てこないユーザーや映画の予測評価値は、平均評価値とする
average_score = dataset.train.rating.mean()
movie_rating_predict = dataset.test.copy()
pred_results = []
for i, row in dataset.test.iterrows():
user_id = row["user_id"]
if user_id not in user_id2index or row["movie_id"] not in movie_id2index:
pred_results.append(average_score)
continue
user_index = user_id2index[row["user_id"]]
movie_index = movie_id2index[row["movie_id"]]
pred_score = pred_matrix[user_index, movie_index]
pred_results.append(pred_score)
movie_rating_predict["rating_pred"] = pred_results
# 各ユーザに対するおすすめ映画は、そのユーザがまだ評価していない映画の中から予測値が高い順にする
pred_user2items = defaultdict(list)
user_evaluated_movies = dataset.train.groupby("user_id").agg({"movie_id": list})["movie_id"].to_dict()
for user_id in dataset.train.user_id.unique():
if user_id not in user_id2index:
continue
user_index = user_id2index[row["user_id"]]
movie_indexes = np.argsort(-pred_matrix[user_index, :])
for movie_index in movie_indexes:
movie_id = user_movie_matrix.columns[movie_index]
if movie_id not in user_evaluated_movies[user_id]:
pred_user2items[user_id].append(movie_id)
if len(pred_user2items[user_id]) == 10:
break
return RecommendResult(movie_rating_predict.rating_pred, pred_user2items)
if __name__ == "__main__":
SVDRecommender().run_sample()