LibRecommender

Overview

LibRecommender is an easy-to-use recommender system focused on end-to-end recommendation. The main features are:

Implement a number of popular recommendation algorithms such as SVD, DeepFM, BPR etc.
A hybrid system, allow user to use either collaborative-filtering or content-based features.
Ease of memory usage, automatically convert categorical features to sparse representation.
Suitable for both explicit and implicit datasets, and negative sampling can be used for implicit dataset.
Making use of Cython or Tensorflow to accelerate model training.
Provide end-to-end workflow, i.e. data handling / preprocessing -> model training -> evaluate -> serving.

Usage

pure collaborative-filtering example :

from libreco.dataset import DatasetPure   # pure data, algorithm svd++
from libreco.algorithms import SVDpp

conf = {
    "data_path": "path/to/your/data",
    "length": "all",
}

dataset = DatasetPure()
dataset.build_dataset(**conf)

svd = SVDpp(n_factors=32, n_epochs=200, lr=0.001, batch_size=4096, task="rating")
svd.fit(dataset, verbose=1)
print(svd.predict(1, 2))	     # predict preference of user 1 to item 2
print(svd.recommend_user(1, 7))	 # recommend 7 items for user 1

include features example :

from libreco.dataset import DatasetFeat   # feat data, algorithm DeepFM
from libreco.algorithms import DeepFmFeat

conf = {
    "data_path": "path/to/your/data",
    "length": 500000,
    "user_col": 0,
    "item_col": 1,
    "label_col": 2,
    "numerical_col": [4],
    "categorical_col": [3, 5, 6, 7, 8],
    "merged_categorical_col": None,
    "user_feature_cols": [3, 4, 5],
    "item_feature_cols": [6, 7, 8],
    "convert_implicit": True,
    "build_negative": True,
    "num_neg": 2,
    "sep": ",",
}

dataset = DatasetFeat(include_features=True)
dataset.build_dataset(**conf)

dfm = DeepFmFeat(lr=0.0002, n_epochs=10000, reg=0.1, embed_size=50,
                 batch_size=2048, dropout_rate=0.0, task="ranking", neg_sampling=True)
dfm.fit(dataset, pre_sampling=False, verbose=1)
print(dfm.predict(1, 10))             # predict preference of user 1 to item 10
print(dfm.recommend_user(1, 7))   # recommend 7 items for user 1

Data Format

JUST normal data format, each line represents a sample. By default, model assumes that user, item, and label column index are 0, 1, and 2, respectively. But you need to specify user, item, and label column index if that’s not the case. For Example, the movielens-1m dataset:

1::1193::5::978300760
1::661::3::978302109
1::914::3::978301968
1::3408::4::978300275

leads to the following settings in conf dict : "user_col": 0, "item_col": 1, "label_col": 2, "sep": "::" .

Besides, if you want to use some other meta features (e.g., age, sex, category etc.), numerical and categorical column index must be assigned. For example, "numerical_col": [4], "categorical_col": [3, 5, 6, 7, 8], which means all features must be in a same table.

Installation & Dependencies

From pypi: pip install LibRecommender

Python 3.5 +
tensorflow >= 1.12
numpy >= 1.15.4
pandas >= 0.23.4
scipy >= 1.2.1
scikit-learn >= 0.20.0

References

Algorithm	Category	Paper
userKNN / itemKNN	pure	Item-Based Collaborative Filtering Recommendation Algorithms
SVD	pure	Matrix Factorization Techniques for Recommender Systems
SVD ++	pure	Factorization Meets the Neighborhood: a Multifaceted Collaborative Filtering Model
superSVD	pure	Factorization Meets the Neighborhood: a Multifaceted Collaborative Filtering Model
ALS	pure	1. Matrix Completion via Alternating Least Square(ALS) / 2. Collaborative Filtering for Implicit Feedback Datasets / 3. Applications of the Conjugate Gradient Method for Implicit Feedback Collaborative Filtering
NCF	pure	Neural Collaborative Filtering
BPR	pure	BPR: Bayesian Personalized Ranking from Implicit Feedback
Wide & Deep	feat	Wide & Deep Learning for Recommender Systems
FM	feat	Factorization Machines
DeepFM	feat	DeepFM: A Factorization-Machine based Neural Network for CTR Prediction
YouTubeRec	feat	Deep Neural Networks for YouTube Recommendations

License

MIT

Metadata

Release files for LibRecommender 0.0.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for LibRecommender 0.0.1
File	Size	Uploaded
LibRecommender-0.0.1.tar.gz	864.9 kB	Oct 28, 2019	Details

Release files / LibRecommender-0.0.1.tar.gz

Download URL	LibRecommender-0.0.1.tar.gz
Size	864.9 kB
Tags	Source
SHA-256 checksum How to use checksums	`5b5b3f6cad3c8493ebadf6bf754f65d3fde910fe1eeaf09d18f9880dc38cbeb9`
BLAKE2b-256 checksum How to use checksums	`a306c80ed1103394ee9e6dc9f31b1ac53ed66d00741f14d0b56acb1cd0d1e824`
Upload date	Oct 28, 2019
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/2.0.0 pkginfo/1.5.0.1 requests/2.22.0 setuptools/41.0.1 requests-toolbelt/0.9.1 tqdm/4.36.1 CPython/3.6.8`