Installing Numeric
==================

As of Numeric 24.0, customizing the installation of Numeric is actually easy.
You'll need

* Python (supposedly 2.0 or greater; only tested for 2.2 or greater).

  If you've installed Python using a package on a Linux system, check that
  you have the appropiate development package (python-devel on rpm-based
  systems such as Mandrake or Red Hat, python-dev on Debian).

* a C compiler. Preferably the same one used to compile Python.

For better performance with linear algebra, you should have optimized
LAPACK libraries, such as ATLAS_. In addition, for faster matrix
multiplication, you'll want a CBLAS library (ATLAS provides one, and
the Gnu Scientific Library can be compiled with an interface to a Fortran
BLAS).

Edit ``customize.py`` to set the appropiate options.

Once you're happy, do::

    $ python setup.py build
    $ python setup.py install

Since Numeric uses the standard Python distutils, you can change where it's
installed by adding ``--prefix=/other/directory`` to the install line.
Alternatively, see `Making a Python Egg`_ below.

Installing on Mac OS X
----------------------

No customizations should be needed. If the ``vecLib`` framework is found
(which it should be for 10.2 and above), it will be used for BLAS and LAPACK.

Installing on Windows
---------------------

If you're using cygwin with the mingw compiler, it is suggested that you
run::

    $ python setup.py config

first. This may fix a bug in LinearAlgebra, where certain routines in
the lapack_lite would be compiled incorrectly.

Making a Python Egg
-------------------

With Numeric 24.2, you can make a `Python egg`_ easily. Simply:

    $ python setup.py bdist_egg

You can then find the egg in the ``dist/`` directory. This can then
be installed using ``easy_install``. Note that ``python setup.py install`` will _not_ install Numeric as an egg.

.. _ATLAS: http://math-atlas.sourceforge.net/
.. _`Python egg`: http://peak.telecommunity.com/DevCenter/PythonEggs