推薦システムの実行

パッケージのインストール

(recsys_full) offline$ pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu124
# ...（3分程度）...

(recsys_full) offline$
 pip install ray
 pip install pyarrow
 pip install kmeans_pytorch
 pip install recbole

(recsys_full) offline$ pip freeze
...（略）...
absl-py==2.5.0
attrs==26.1.0
click==8.5.0
colorama==0.4.4
colorlog==4.7.2
cuda-bindings==13.3.1
cuda-pathfinder==1.8.0
cuda-toolkit==13.0.3.0
filelock==3.32.4
fsspec==2026.7.0
grpcio==1.83.1
Jinja2==3.1.6
jsonschema==4.26.0
jsonschema-specifications==2025.9.1
kmeans-pytorch==0.3
Markdown==3.10.3
MarkupSafe==3.0.3
mpmath==1.3.0
msgpack==1.2.2
networkx==3.6.1
nvidia-cublas==13.1.1.3
nvidia-cuda-cupti==13.0.85
nvidia-cuda-nvrtc==13.0.88
nvidia-cuda-runtime==13.0.96
nvidia-cudnn-cu13==9.20.0.48
nvidia-cufft==12.0.0.61
nvidia-cufile==1.15.1.6
nvidia-curand==10.4.0.35
nvidia-cusolver==12.0.4.66
nvidia-cusparse==12.6.3.3
nvidia-cusparselt-cu13==0.8.1
nvidia-nccl-cu13==2.29.7
nvidia-nvjitlink==13.3.33
nvidia-nvshmem-cu13==3.4.5
nvidia-nvtx==13.0.85
plotly==7.0.0
protobuf==7.36.0
pyarrow==25.0.1
PyYAML==6.0.3
ray==2.58.0
recbole==1.2.0
referencing==0.37.0
rpds-py==2026.6.3
setuptools==84.0.0
sympy==1.14.0
tabulate==0.10.0
tensorboard==2.21.0
tensorboard-data-server==0.7.2
texttable==1.7.0
thop==0.1.1.post2209072238
torch==2.13.0
torchaudio==2.11.0
torchvision==0.28.0
triton==3.7.1
typing_extensions==4.16.0
Werkzeug==3.1.8
...（略）...

推薦処理プログラムの準備

下記ファイルをダウンロードし、offline/src/ディレクトリに移動する。
- recsyslab / recsys-full / src / offline / src /
  - recommenders/__init__.py
  - recommenders/base_recommender.py
  - recommenders/bpr_recommender.py
  - recommenders/movie_similarity_recommender.py
  - recommenders/pupularity_recommender.py
  - utils/__init__.py
  - utils/dataset.py
  - utils/recommend_result.py
  - config.yaml
  - offline.ini
  - update.py

(recsys_full) offline$ tree src/
src/
├── config.yaml                           # <-
├── keygen.py
├── ml2rdb.py
├── offline.ini                           # <-
├── recommenders                          # <-
│   ├── __init__.py                      # <-
│   ├── base_recommender.py              # <-
│   ├── bpr_recommender.py               # <-
│   ├── movie_similarity_recommender.py  # <-
│   └── popularity_recommender.py        # <-
├── update.py                             # <-
├── user2encrypt.py
└── utils                                 # <-
    ├── __init__.py                       # <-
    ├── dataset.py                        # <-
    └── recommend_result.py               # <-

3 directories, 14 files

推薦システムの実行

(recsys_full) offline$ python src/update.py --ini src/offline.ini
processing Dataset.to_rating_matrix: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 100836/100836 [00:20<00:00, 4972.83it/s]
...（略）...
processing Dataset.to_movies_movies_similarity_matrix: 100%|████████████████████████████████████████████████████████████████████████████████████| 9742/9742 [18:20<00:00,  8.86it/s]
processing PopularityRecommender.recommend: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:00<00:00, 165.79it/s]
processing MovieSimilarityRecommender.recommend: 100%|████████████████████████████████████████████████████████████████████████████████████████| 9742/9742 [00:03<00:00, 2769.41it/s]
...（略）...
processing BPRRecommender.recommend: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████| 611/611 [00:02<00:00, 257.93it/s]
elapsed_time:1343.605[sec]
# ...（25分程度）...

結果の確認

(recsys_full) offline$ tree data/
data/
├── genres.csv
├── movies.csv
├── movies_genres.csv
├── ratings.csv
├── reclist_bpr.csv               # <-
├── reclist_movie_similarity.csv  # <-
├── reclist_popularity.csv        # <-
├── tags.csv
└── users.csv

1 directory, 9 files

(recsys_full) offline$ tree local/
local/
├── keys
├── ml-latest-small
│   ├── README.txt
│   ├── links.csv
│   ├── movies.csv
│   ├── ratings.csv
│   └── tags.csv
├── ml-ls.inter                          # <-
├── ml-ls.item                           # <-
├── ml-ls.user                           # <-
├── ml-rdb
│   ├── genres.csv
│   ├── links.csv
│   ├── movies.csv
│   ├── movies_genres.csv
│   ├── ratings.csv
│   ├── tags.csv
│   └── users.csv
├── movies_movies_similarity_matrix.csv  # <-
├── rating_matrix.csv                    # <-
└── users_.csv

3 directories, 19 files

(recsys_full) offline$
 less data/reclist_popularity.csv
 less data/reclist_movie_similarity.csv
 less data/reclist_bpr.csv

参考

風間正弘，飯塚洸二郎，松村優也，『著推薦システム実践入門 ―仕事で使える導入ガイド』，オライリー・ジャパン，2022．
与謝秀作，『特集 3 最新レコメンドエンジン総実装協調フィルタリングから深層学習まで』，WEB+DB PRESS Vol.129，pp.69-100，技術評論社，2022．
[🐛BUG] Wrong number of users and items in datasets information · Issue #1516 · RUCAIBox/RecBole
RecBole v1.2.0 — RecBole 1.2.0 documentation