Sommelier is an experimental recommender system written in Python and the Flask framework, with a simple JSON API backed up by a MySQL database. It implements several methods for generating recommendations based on collaborative filtering and matrix factorization.

Sommelier is not used in production anywhere - it was my undergraduate project at uni, so it's just on GitHub for academic purposes really. There are much better recommender systems out there to crib from, but feel free to poke around the code and ask me any questions if you're interested.

We need pip to be installed to manage python packages

sudo apt-get install python-pip
pip install -U pip

Vitualenv enables us to install modules and packages local to our project, so we dont need to expose our system-level python installation to incompatible or otherwise obnoxious packages that might destabilize our other project, or OS in general

pip install virtualenv
virtualenv flask

MySQL-python will enable us to query data (very useful!) - helpful advice retrieved from http://codeinthehole.com/writing/how-to-set-up-mysql-for-python-on-ubuntu/

There are some additional dependencies with MySQL-python that need to be installed at a system level

sudo apt-get install libmysqlclient-dev python-dev

Documentation on working with MySQLdb is at: http://mysql-python.sourceforge.net/MySQLdb.html

Now we can install MySQL-python in our virtualenv using a local pip

./flask/bin/pip install MySQL-python

Flask itself is our web framework of choice

./flask/bin/pip install flask

Mock will be useful for unit testing

./flask/bin/pip install mock

nose will serve as our test runner https://nose.readthedocs.org/en/latest/ Source: https://github.com/nose-devs/nose

./flask/bin/pip install nose

Scipy is a pain. The following installation method is courtesy of: http://www.scipy.org/Installing_SciPy/Linux#head-d437bf93b9d428c6efeb08575f631ddf398374ea

This installs rather a lot of stuff, the dependencies of NumPy, globally :-|

sudo apt-get build-dep python-numpy 

The following command does a big build and throws all sorts of errors, which are apparently fine to ignore.

sudo apt-get -b source python-numpy 
./flask/bin/pip install scipy 

Install dependencies for python-recsys

flask/bin/pip install csc-pysparse networkx divisi2

Clone python-recsys from Git and set it up in virtualenv

git clone http://github.com/ocelma/python-recsys
cd python-recsys
../flask/bin/python setup.py install

Install matplotlib for graphing test results.

N.b. this is not subsequently installed in the virtualenv, but that isnt a problem as the graphing can be done without invoking any of our virtualenv

sudo apt-get install python-matplotlib

To run the API:

flask/bin/python app.py

This will fire up Flask on localhost:5000

To run tests:

flask/bin/nosetests

These fail at the moment, because I broke some things at the last moment before handing my project in. Because I'm an idiot!