Highly optimized and parallelized methods for Bayesian Factorization, including BPMF and Macau. The package uses optimized OpenMP/C++ code with a Cython wrapper to factorize large scale matrices. Macau method is able to perform matrix and tensor factorization while incorporating high-dimensional side information to the factorization.

For examples see documentation.

To install Macau it possible to use pre-compiled binaries or compile it from source.

# install dependencies:
sudo apt-get install python-pip python-numpy python-scipy python-pandas cython
sudo apt-get install libopenblas-dev autoconf gfortran

# checkout and install Macau
git clone https://github.com/jaak-s/macau.git
cd macau
python setup.py install --user

Make sure that pip is available and OpenBLAS or some other BLAS is installed. Then install following packages using pip: numpy scipy pandas cython, e.g. by

# skip any of them if already available
pip install numpy --user
pip install scipy --user
pip install pandas --user
pip install cython --user

# checkout and install Macau
git clone https://github.com/jaak-s/macau.git
cd macau
python setup.py install --user

# install dependencies
pip install numpy
pip install scipy
pip install pandas
pip install cython
# install brew (http://brew.sh/)
brew install homebrew/science/openblas
brew install gcc

# checkout and install Macau
git clone https://github.com/jaak-s/macau.git
cd macau
CC=g++-5 CXX=g++-5 python setup.py install

Macau is also available using Docker image at stadius/macau.

Without mounting a local directory the docker can be executed by

docker run -it --rm -p 8888:8888 stadius/macau

To mount a local directory add -v ~/my_data_dir:/data where ~/my_data_dir is on the local system and /data will be the folder in the container:

docker run -v ~/my_data_dir:/data -it --rm -p 8888:8888 stadius/macau

There is a plan to support Python wheel packages. Currently, we do not have one built yet.

• Jaak Simm (Macau C++ version, Cython wrapper, Macau MPI version, Tensor factorization)
• Adam Arany (Probit noise model)
• Tom Vander Aa (OpenMP optimized BPMF)
• Tom Haber (Original BPMF code)