[SciPy-User] Completing source install of numpy/scipy

[David]

On 05/02/2011 05:23 PM, Glen Shennan wrote:
> Hi,
>
> This question may be slightly outside the scope of the list but I have
> been unable to find a solution myself and no forums have provided answers.
>
> I'm trying to install the standard python/numpy/scipy/matplotlib
> combination.  I'm using Ubuntu 11.04 and am using python 2.7 obtained
> from the repos.  I have numpy and scipy (and BLAS/ATLAS/LAPACK) compiled
> from source and installed in $HOME/local/[lib|include|bin]/
>
> I now want to install matplotlib.  I'm happy to do this from the repos,
> I only want the number-crunching parts from source for speed.  However
> matplotlib depends on the packages python-numpy, libblas3gf and
> liblapack, so whenever I try to install python-matplotlib these packages
> are installed as well.  I've tried "aptitude hold ..." and locking the
> version in synaptic but for reasons that are beyond me the packages
> continue to be installed no matter what I do.
>
> I realise I could just install matplotlib too but I run more than a
> couple of other programs that have the same dependencies so I need a
> general solution.
>
> How might I go about convincing Linux that the packages are installed?

This is impossible.

If you want to use ubuntu packages which depend on numpy, you have to 
install numpy from ubuntu. What you *can* do is to build your own 
numpy/scipy and make sure it still works with the ubuntu matplotlib 
package (i.e. your custom numpy will be used). Still not so easy, but 
doable.

But if you only care about "speed", then using the ubuntu packages 
should work relatively well, unless you want to build a multithreaded 
blas/lapack by yourself, which is not for the faint of the heart.

cheers,

David