This is a step-by-step installation guide to setup a scientific python environment on Mac OS X 10.9+ Mavericks / Yosemite with homebrew.
An older version of this setup guide can be found here: Scientific Python on Mac OS X 10.8 with homebrew, main changes: rename of a tap + changes wrt. openblas as Accelerate was fixed in OS X 10.9
Needless to say: Make a backup (Timemachine)
First install homebrew.
Follow their instructions, then come back here.
If you don’t have a clean install, some of the following steps might need minor additional attention (like changing permissions
chgrp or overwriting existing files in the linking step with
brew link --overwrite package_that_failed. In this case i can only recommend a backup again).
In general: execute the following commands one line at a time and read the outputs! If you read some warnings about “keg-only” that’s fine, it just means that brew won’t “hide” your system’s stuff behind the stuff it installed itself so it doesn’t cause problems… brewed stuff will still use it.
brew tap homebrew/science # a lot of cool formulae for scientific tools
brew tap homebrew/python # numpy, scipy, matplotlib, ...
brew update && brew upgrade
# install a brewed python
brew install python
A word about brewed python: this is what you want!
It’s more up to date than the system python, it will come with pip and correctly install in the brew directories, working together well with brewed python libs that aren’t installable with plain pip. This also means that pip by default will work without sudo as all of homebrew, so if you ever run or have to run
sudo pip ... because of missing permissions, then you’re doing it wrong! Also, don’t be afraid of multiple pythons on your system: you have them anyhow (python2 and python3) and it’s an advantage, as we’ll make sure that nothing poisons your system python and that you as a user & developer will use the brewed python:
# should say /usr/local/bin/python
# /usr/local/bin should appear in front of /usr/bin
If this is not the case you’d probably end up not using brewed python. Please check your brew install with
brew doctor, it will probably tell you that you should consider updating your paths in
~/.bashrc. You can either follow its directions or create a
~/.profile file like this one: ~/.profile. If you performed these steps, please close your terminal windows and open a new one for the changes to take effect. Test the above again.
Even if the above check worked, run the following anyhow and read through its output (no output is good):
Pay special attention if this tells you to install XQuartz, and if it does, install it! You’ll need it anyhow…
Now after these preparations, let’s really start installing stuff… below you’ll mostly find one package / lib per line. For each of them and for their possible options: they’re a recommendation that might save you some trouble, so i’d recommend to install all of them as i write here, even if specifying some of the options will compile brew packages from source and take a bit longer…
# image generating stuff
brew install libtiff libjpeg webp little-cms2
pip install Pillow
brew install imagemagick --with-fftw --with-librsvg --with-x11
brew install graphviz --with-librsvg --with-x11
brew install cairo
brew install py2cairo # this will ask you to download xquartz and install it
brew install qt pyqt
# install virtualenv, nose (unittests & doctests on steroids)
pip install virtualenv
pip install nose
# install numpy and scipy
# there are two ways to install numpy and scipy now: via pip or via brew.
# PICK ONE, i prefer pip for proper virtualenv support and more up-to-date versions.
pip install numpy
pip install scipy
# (if you want to run numpy and scipy with openblas also remove comments below:)
#brew install openblas
brew install numpy # --with-openblas
brew install scipy # --with-openblas
# test the numpy & scipy install
python -c 'import numpy ; numpy.test();'
python -c 'import scipy ; scipy.test();'
# some cool python libs (if you don't know them, look them up)
# matplotlib: generate plots
# pandas: time series stuff
# nltk: natural language toolkit
# sympy: symbolic maths in python
# q: fancy debugging output
# snakeviz: cool visualization of profiling output (aka what's taking so long?)
#brew install Caskroom/cask/mactex # if you want to install matplotlib with tex support and don't have mactex installed already
brew install matplotlib --with-cairo --with-tex # cairo: png ps pdf svg filetypes, tex: tex fonts & formula in plots
pip install pandas
pip install nltk
pip install sympy
pip install q
pip install snakeviz
# ipython with parallel and notebook support
brew install zmq
pip install ipython[all]
# html stuff (parsing)
pip install html5lib cssselect pyquery lxml BeautifulSoup
# webapps / apis (choose what you like)
pip install Flask Django tornado
# semantic web stuff: rdf & sparql
pip install rdflib SPARQLWrapper
# graphs (graph metrics, social network analysis, layouting)
pip install networkx
brew install graph-tool
# maintenance: updating pip libs
pip install pip-tools # you'll then have a pip-review command, see Updating section below
Have fun 😉
OK, let’s say it’s been a while since you installed things with this guide and you now want to update all the installed libs. To do this you should first upgrade everything installed with brew like this:
Afterwards for upgrading pip packages i recommend the pip-tools package:
Once installed you should be able to run the following either in a virtualenv or globally for your whole system:
It will check your installed packages for new versions and give you a list of outdated packages. I’d recommend to run it with the
-i option to interactively install the upgrades. A word of warning about the brewed packages: If i recommended to install a package with brew above that’s usually for a good reason like the pip version not working properly. If you’re a bit more advanced, you can try to upgrade them with pip, but i’d recommend to properly unlink them with
brew unlink <package> before, as some pip packages might run into problems otherwise. If you find the pip package works like a charm then, please let me know in the comments below so i can update this guide. In general i prefer the pip packages as they’re more up to date, work in virtual environments and can then easily be updated with
As always: If you liked this, think something is missing or wrong leave a comment.
Updates to this guide:
2014-03-02: include checking of
$PATH for Mike
2015-03-17: enhanced many explanations, provided some useful options for packages, general workover
2015-04-15: included comment about installing mactex via cask if not there already, thanks to William
2015-06-05: Pillow via pip and Updating section