Tophat is a factorization-based recommendation engine built using TensorFlow.

Consider this a dev fork of https://github.com/saksdirect/tophat . Except this is not actually a fork anymore.

Installing from PyPi:

pip install top-hat

Installing the master branch from github in development mode, run:

git clone git@github.com:gilt/tophat.git
cd tophat
pip install -e .

Note that by default, installation assumes you already have TensorFlow installed. However, if you need, you can include the installation of TensorFlow in the setup extras as following: (choose the one that's right for you)

# CPU pypi
pip install top-hat[tf]

# GPU pypi
pip install top-hat[tf_gpu]

# CPU local dev
pip install -e .[tf]

# GPU local dev
pip install -e .[tf_gpu]

There are two provided Dockerfiles: Dockerfile and Dockerfile.gpu. The latter gpu variant requires nvidia-docker.

There are some minimal tests in tests/ which can all be run using pytest or python setup.py test.

The initial motivation behind tophat was to port over LightFM and Spotlight into TensorFlow.

There also are many other amazing recommender systems out there -- so choose the one that is right for your case.