Notes on installing Numpy/Scipy and friends on a Ubundu

Computer: Mobile Pentium 4, 2.2GHz

OS: Ubuntu 5.10, Linux kernel 2.6.12-10-386

Python: 2.4.2

Local setup to simplify installing python modules as an ordinary user:

# chown -R pearu:admin /usr/local/

Prerequisities

Minimal set of software packages for building numpy:

python-dev gcc

Minimal set of software packages for building scipy:

python-dev gcc g77 g++

Additional recommended software packages for getting higher performance in numpy/scipy components:

atlas3-base-dev atlas3-sse2

Check using

python numpy/distutils/cpuinfo.py

for 3dnow, sse, sse2 support in your CPU.

Building and testing NumPy

$ cd svn/numpy
$ python setup.py install --prefix=/usr/local
$ cd      # get out of numpy source directory
$ python
>>> import numpy
>>> numpy.test(10)
...
Ran 354 tests in 2.302s
...
OK

To rebuild numpy, run

$ cd svn/numpy
$ rm -rf build
$ python setup.py install --prefix=/usr/local

Building and testing SciPy

Optional packages for building SciPy packages:

libumfpack4-dev swig
fftw3-dev

Note: libumfpack4 is linked against atlas2 libraries.

To install scipy, run

$ cd svn/scipy
$ python setup.py install --prefix=/usr/local
$ cd # get out of scipy source directory
$ python
>>> import scipy
>>> scipy.test(10)
...
Ran 1520 tests in 65.730s
...
OK

Installing matplotlib

Building matplotlib requires the following packages to be installed:

libglib2.0-dev libgtk2.0-dev python-gtk2-dev tk8.4-dev dvipng

Make sure that you can open display from a command prompt, e.g. use ssh -X login, before trying to build matplotlib.

Get matplotlib from svn:

$ svn co https://svn.sourceforge.net/svnroot/matplotlib/trunk/matplotlib matplotlib

To install and test matplotlib, run

$ cd svn/matplotlib
$ python setup.py install --prefix=/usr/local
$ cd
$ export NUMERIX=numpy
$ python
>>> from pylab import *
>>> plot([1,2])
>>> show()

Installing ipython

Get ipython from svn:

$ svn co http://ipython.scipy.org/svn/ipython/ipython/trunk ipython

$ cd svn/ipython
$ python setup.py install --prefix=/usr/local
$ cd
$ ipython -pylab
>>> plot([1,2])

Installing Enthought (numpy port)

To install enthought, you must have the following packages installed:

python-vtk libxtst-dev

To use enthought, have the following packages installed:

python-wxgtk2.6 python-celementtree msttcorefonts

To install and test enthought, run

$ cd svn/enthought_lib_numpy
$ python setup.py install --prefix=/usr/local
$ cd
$ python
>>> import enthought
>>> enthought.test()