Unfortunately, Linux distributions rarely support install as non-root, so your only solution is almost always to install from sources. For python >= 2.6, the easiest solution is the --user option:

python setup.py install --user

This will install numpy in $HOME/.local, and python automatically looks there for packages. You're done.

For python < 2.6, you may recreate a local hierarchy in your home directory (/home/joe/local/bin for binaries, /home/joe/local/lib for libraries, etc...), and install numpy in this hierarchy:

python setup.py install --prefix=$HOME/local

In that case, you need to tell python to look in $HOME/local, though. This can be controled by the PYTHONPATH variable. If your shell is bash (the default shell on most linux distributions), add this line in your .bashrc:

export PYTHONPATH=$HOME/local/lib/python2.5/site-packages:$PYTHONPATH

(of course, adapt the python version if necessary). Then, numpy will be available in any newly created shell, i.e.:

python -c "import numpy; print numpy.__file__"

should work and print something like /home/joe/local/lib/python2.5/site-packages/numpy/init.pyc.