Installing iPython, Numpy, Scipy, and Matplotlib on Leopard

[EDIT: This is pretty old now, it probably works if you want a whole separate install, but I've recently obtained some new macs and found it can be much easier. Details in this post]

So after upgrading to Leopard I found my Python plotting stuff didn’t work, so had to reinstall it. My last setup was decidedly dodgy, being something of a mix of 2.4 and 2.5, which wouldn’t work if I removed one. Very suspect.

I have a MacBook Pro, recently upgraded to Leopard. Mostly I followed these instructions, but there were some steps skipped that I had to include for my particular case. Possibly because I’m on an intel macbook and the steps were for a ppc.

[I recently noticed this page, which you might want to read first since you might not really have to do a whole new install]

First thing I needed to do was get the latest XCode (3). A 1Gb download since my X DVD was at home. I also got the latest gfortran from here (gfortran-4.2.1.dmg) and installed fftw as decribed on the scipy pages.

I started totally from scratch, removing /Library/Frameworks/Python.framework/ , MacPython from Applications, and the executable links in /usr/local/bin . I also got rid of the other stuff I’d installed in /Library/Python/2.X/site-packages .

I first got the MacPython 2.5.1 install and put that on. You get it from here, and it’s a simple click and install thing. Done. It gave me some thing about a postflight error, which apparently adds the Python.framework to my path, but it didn’t seem to matter.

Next was iPython from here. I got version 0.8.2. This install went fine. Checked by running ipython in a terminal. Most of the python things are installed by typing

sudo python setup.py install

Next was getting numpy and scipy from here. I installed numpy first no worries, and tested it with “import numpy” and then “numpy.test(1,10)” within the ipython shell. Scipy installed easily too. Like the guy I copied, I get one failure when running the test “import scipy” and “scipy.test(1,10)”. No worries.

Finally it came to matplotlib, from here. At this point I followed the steps here, and some other bits I found. There was a problem where the python I was trying to install with wouldn’t deal with a ppc/intel universal compile, and I had to use /System/Library/Frameworks/Python.framework/Versions/Current/bin/python instead of the default in the /Library framework. After finishing successfully, I had to manually copy the dateutil directory to /Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/ for some reason.

Finally I had to get wxPython from here, and do the click install.

After all that it finally works. I can’t really say I’m an expert since it took me a day to figure all that out. I think it’s worth it for how great the matplotlib interface is when it works though. Where else can you make nice plots that are draggable, zoomable, customisable so easily?