Nix for Data Science

As Ubuntu's LTS releases usually do not ship the latest versions, it becomes challenging to set up a clean environment for Data Science. Nix provides us with an easy-to-use alternative without needing to interfere with the host system.

Here, we build a Python 3.5 environment with packages commonly used for Data Science. The packages are defined in default.nix instead of using Python's package manager (pip).

Nix

First off, install Nix:

$ bash <(curl https://nixos.org/nix/install)
$ source ~/.nix-profile/etc/profile.d/nix.sh

Building the environment

Clone this repository, change the working directory to it and run nix-build:

$ git clone git@github.com:tindzk/nix-ds.git
$ cd nix-ds
$ nix-build

Now, instead of using the host's Python tools, you can use the ones from result/bin/:

$ result/bin/python3
$ result/bin/ipython notebook

Upgrading environment

The environment can be easily upgraded:

$ nix-channel --update
$ nix-env -u

See also

• Getting started with Nix package manager
• Nix for Python developers