wujinming / surprise

A Python scikit for building and analyzing recommender systems

Surprise

Overview

Surprise is a Python scikit building and analyzing recommender systems.

Surprise was designed with the following purposes in mind:

• Give the user perfect control over his experiments. To this end, a strong emphasis is laid on documentation, which we have tried to make as clear and precise as possible by pointing out every details of the algorithms.
• Alleviate the pain of Dataset handling. Users can use both built-in datasets (Movielens, Jester), and their own custom datasets.
• Provide various ready-to-use prediction algorithms such as baseline algorithms, neighborhood methods, matrix factorization-based ( SVD, PMF, SVD++, NMF), and many others. Also, various similarity measures (cosine, MSD, pearson...) are built-in.
• Make it easy to implement new algorithm ideas.
• Provide tools to evaluate, analyse and compare the algorithms performance. Cross-validation procedures can be run very easily, as well as exhaustive search over a set of parameters.

The name SurPRISE (roughly :) ) stands for Simple Python RecommendatIon System Engine.

Getting started, example

Here is a simple example showing how you can (down)load a dataset, split it for 3-folds cross-validation, and compute the MAE and RMSE of the SVD algorithm.

from surprise import SVD
from surprise import Dataset
from surprise import evaluate, print_perf


# Load the movielens-100k dataset (download it if needed),
# and split it into 3 folds for cross-validation.
data = Dataset.load_builtin('ml-100k')
data.split(n_folds=3)

# We'll use the famous SVD algorithm.
algo = SVD()

# Evaluate performances of our algorithm on the dataset.
perf = evaluate(algo, data, measures=['RMSE', 'MAE'])

print_perf(perf)

Output:

Evaluating RMSE, MAE of algorithm SVD.

        Fold 1  Fold 2  Fold 3  Mean
MAE     0.7475  0.7447  0.7425  0.7449
RMSE    0.9461  0.9436  0.9425  0.9441

Surprise can do much more (e.g, GridSearch)! You'll find more usage examples in the documentation .

Benchmarks

Here are the average RMSE, MAE and total execution time of various algorithms (with their default parameters) on a 5-folds cross-validation procedure. The datasets are the Movielens 100k and 1M datasets. The folds are the same for all the algorithms (the random seed is set to 0). All experiments are run on a small laptop with Intel Core i3 1.7 GHz, 4Go RAM. The execution time is the real execution time, as returned by the GNU time command.

Movielens 100k	RMSE	MAE	Time (s)
NormalPredictor	1.5228	1.2242	4
BaselineOnly	.9445	.7488	5
KNNBasic	.9789	.7732	27
KNNWithMeans	.9514	.7500	30
KNNBaseline	.9306	.7334	44
SVD	.9396	.7412	46
SVD++	.9200	.7253	31min
NMF	.9634	.7572	55
Slope One	.9454	.7430	25
Co clustering	.9678	.7579	15

Movielens 1M	RMSE	MAE	Time (min)
NormalPredictor	1.5037	1.2051	< 1
BaselineOnly	.9086	.7194	< 1
KNNBasic	.9207	.7250	22
KNNWithMeans	.9292	.7386	22
KNNBaseline	.8949	.7063	44
SVD	.8936	.7057	7
NMF	.9155	.7232	9
Slope One	.9065	.7144	8
Co clustering	.9155	.7174	2

Installation / Usage

The easiest way is to use pip (you'll need numpy):

$ pip install numpy
$ pip install scikit-surprise

Or you can clone the repo and build the source (you'll need Cython and numpy):

$ git clone https://github.com/NicolasHug/surprise.git
$ python setup.py install

License

This project is licensed under the BSD 3-Clause license, so it can be used for pretty much everything, including commercial applications. Please let us know how Surprise is useful to you!

Acknowledgements:

• Pierre-François Gimenez, for his valuable insights on software design.
• Maher Malaeb, for the GridSearch implementation.

Contributing, feedback, contact

Any kind of feedback/criticism would be greatly appreciated (software design, documentation, improvement ideas, spelling mistakes, etc...).

If you'd like to see some features or algorithms implemented in Surprise, please let us know! Some of the current ideas are:

• Bayesian PMF
• RBM for CF

Please feel free to contribute (see guidelines) and send pull requests!

To contact us, send us a tweet or mail us at contact at nicolas-hug dot com. For any bug or issue, please use the GitHub project page.