A library implementing Bayesian Personalised Ranking (BPR) for Matrix Factorisation, as described by Rendle et al. in :
http://arxiv.org/abs/1205.2618
This model tries to predict a personalised ranking of items from a user's viewing history. It has been shown to be very efficient for recommendation tasks. It's also used in a variety of other tasks, such as matrix completion, link prediction and tag recommendation.
This library uses Theano and can therefore run on a GPU through CUDA or on the CPU, for which you'll need a working BLAS. We recommend using OpenBlas.
$ pip install theano-bpr
An iPython Notebook demonstrating the use of theano-bpr over the Movielens dataset is available in examples/.
$ nosetests
See 'COPYING' and 'AUTHORS' files