Skip to content
This repository was archived by the owner on Nov 2, 2018. It is now read-only.

pythonbrew_ipython.rst

Latest commit

 

History

History
256 lines (175 loc) · 7.14 KB

pythonbrew_ipython.rst

File metadata and controls

256 lines (175 loc) · 7.14 KB

Using pythonbrew to install minimum scientific computing environment on OS X 10.8

Here is a list of packages that we want to install

$ pip freeze
distribute==0.6.28
h5py==2.0.1
ipython==0.13
matplotlib==1.1.1
numpy==1.6.2
pyzmq==2.2.0.1
scipy==0.10.1
tornado==2.4
wsgiref==0.1.2

First Step: Install Xcode and gfortran

First, you have should have Xcode 4.4.1 installed so you can compile some C program.

You will need the "command line tools" from Xcode 4.4.1. A version of such tool can be downloaded from https://developer.apple.com/downloads/index.action (an Apple ID is needed), or using Xcode 4.4.1 through the "Download Pane" in the preference setting to download and install it. You will also need some Fortran complier to install scipy too.

There are various ways to install the fortran compiler with OS X 10.8 as shown in this page http://hpc.sourceforge.net. Among the different ways to install Fortran, the easiest one is to download gcc-mlion.tar.gz (http://prdownloads.sourceforge.net/hpc/gcc-mlion.tar.gz) and directly un-tar it to install gcc 4.8 + gfortran in to /usr/local/. However, this might cause some problem as the package does include the gcc 4.8 compiler which may cause some confliction with the Xcode's gcc compiler. Here is the order of directory for binary lookup in my PATH environment variable:

/usr/bin:/bin:/usr/sbin:/sbin:/usr/local/bin:/Applications/Xcode.app/Contents/Developer/usr/bin/

It will allow to use the gcc from Xcode and gfortran from /usr/local/bin.

Install pythonbrew

One can find the installation guide for pythonbrew here https://github.com/utahta/pythonbrew

The installation on OS X 10.8 is straight forward. We just need to follow the instruction from the pythonbrew's github page. Open a terminal and use the following command

$ curl -kL http://xrl.us/pythonbrewinstall | bash

If you don't want to invoke pythobrew automatically, you don't need add this line to the .bashrc

[[ -s "$HOME/.pythonbrew/etc/bashrc" ]] && source "$HOME/.pythonbrew/etc/bashrc"

Instead, you can load the pythonbrew environment within bash by

$ source "$HOME/.pythonbrew/etc/bashrc"

to invoke pythobrew manually. This is my preferred way to make sure I am aware what environment I am using.

Now, install python 2.7.2

$ pythonbrew install 2.7.2

At this stage, you can use the pythonbrew's python by asking pythonbrew to setup the environment for you

$ pythonbrew use Python-2.7.2

You have a clean environment now. You can check the python package installed by

$ pip freeze
distribute==0.6.28
wsgiref==0.1.2

Install numpy

There is a potential bug in pip such that the direct commend pip install numpy might not install everything that we need for installing scipy. If you install numpy by

$ pip install numpy

It will miss a configuration file (npymath.ini) that is needed for compiling scipy. Instead, we can download the numpy source using pip

$ pip install  numpy -d work_dir
$ cd work_dir
$ unzip numpy-numpy-1.6.2.zip

then install numpy with setup.py

$ cd numpy-numpy-1.6.2
$ python setup.py install

You should check if npymath.ini exists

$ cat /Users/username/.pythonbrew/pythons/Python-2.7.2/lib/python2.7/site-packages/numpy/core/lib/npy-pkg-config/npymath.ini
[meta]
Name=npymath
Description=Portable, core math library implementing C99 standard
Version=0.1

[variables]
pkgname=numpy.core
prefix=${pkgdir}
libdir=${prefix}/lib
includedir=${prefix}/include

[default]
Libs=-L${libdir} -lnpymath
Cflags=-I${includedir}
Requires=mlib

[msvc]
Libs=/LIBPATH:${libdir} npymath.lib
Cflags=/INCLUDE:${includedir}
Requires=mlib

Install scipy

Similarly, it is better to download the source of scipy to install it with setup.py

$ pip install scipy -d work_dir
$ cd work_dir
$ unzip scipy-0.10.1.zip
$ cd scipy-0.10.1

However, due to the C header changes in OS X 10.8, we will need to modify a number of lines of the C source code in scipy to be able to compile it under 10.8. This is the kind of errors that one will encounter if we use python setup.py to install scipy

/System/Library/Frameworks/vecLib.framework/Headers/vecLib.h:22:4: error: #error "<vecLib/vecLib.h> is deprecated.  Please #include <Accelerate/Accelerate.h> and link to Accelerate.framework."

One way to bypass such error is the set the __ACCELERATE__ compiler flag

$ export CFLAGS=-D__ACCELERATE__

Once the flag is set, use setup.py to install scipy

$ cd work_dir/scipy
$ python setup.py install

At this stage, you should have $numpy$ and $scipy$ installed

$ pip freeze
distribute==0.6.28
numpy==1.6.2
scipy==0.10.1
wsgiref==0.1.2

Install libhdf5

Download libhdf5 from http://www.hdfgroup.org/ftp/HDF5/current/src/, un-tar the downloaded hdf5-1.8.9.tar.bz2

$ tar jxvf hdf5-1.8.9.tar.bz2

Configure and build libhdf5

$ cd hdf5-1.8.9
$ ./configure --prefix=$HOME/.pythonbrew/ --enable-cxx
$ make
$ make install

Install h5py

Download h5py use pip

$ pip install h5py -d work_dir
$ cd work_dir
$ tar zxvf h5py-2.0.1.tar.gz
$ cd zxvf h5py-2.0.1

Install h5py, the only thing one need to worry about is to tell setup.py where the libhdf5's root directory is.

$ python setup.py build --hdf5=$HOME/.pythonbrew
$ python setup.py install

Install IPython

We need tornado and pyzmq so the cool ipythno notebook can work

$ pip install tornado
$ pip install pyzmq

Install IPython

$ pip install ipython

Now, we get these packages

$ pip freeze ipython
distribute==0.6.28
h5py==2.0.1
ipython==0.13
numpy==1.6.2
pyzmq==2.2.0.1
scipy==0.10.1
tornado==2.4
wsgiref==0.1.2

Install matplotlib

prerequisites: freetype2 and libpng

install freetype2

$ mv ~/Downloads/freetype-2.4.10.tar.bz2 .
$ tar jxvf freetype-2.4.10.tar.bz2
$ cd freetype-2.4.10
$ ./configure --prefix=$HOME/.pythonbrew/
$ make install

install libpng

$ wget http://ncu.dl.sourceforge.net/project/libpng/libpng15/1.5.12/libpng-1.5.12.tar.gz
$ tar zxvf libpng-1.5.12.tar.gz
$ cd ./lpng1512/
$ rm -rf ~/Downloads/lpng1512-1/
$ cp scripts/makefile.darwin makefile
$ vi makefile  # change the prefix variable, point it to the $HOME/.pythonbrew directory
$ make install

install matplotlib

$ pip install matplotlib -d work_dir
$ cd work_dir/
$ tar zxvf matplotlib-1.1.1.tar.gz
$ cd matplotlib-1.1.1
$ python setup.py install