Generally, operations does not want us to use pip & pypi in production. One alterative is to distribute dependencies as a set of python eggs. However, installing a package via an egg results in arbitrary code execution. Recently, we have been considering using Debian files to capture our dependencies, but that proved to be overly complex. Python wheels (binary distribution) provide an alternative to eggs that does not require arbitrary code execution. So, it would be great if we could use wheels instead.
This task is done when someone tries to build all of the [dependencies for running ores-wikimedia-config](https://github.com/wiki-ai/ores-wikimedia-config/blob/master/requirements.txt) and reports how it works out.
Python wheels: http://pythonwheels.com/
Note that building wheels is trivial for pure-python libraries. How hard is it to generate wheels for sklearn, scipy, numpy, etc?