[SciPy-user] Install scipy on ubnutu 8.10
Sun Oct 5 15:15:35 CDT 2008
Nathan Bell wrote:
> On Sun, Oct 5, 2008 at 3:01 PM, Xavier Gnata <firstname.lastname@example.org> wrote:
>> I have installed an ubuntu 8.10 within a kvm image.
>> The goal is to write an as simple as possible procedure to compile
>> scipy on this distribution.
>> I would like to be able to compile scipy only after having installed
>> some packages. No tricks.
> Hi Xavier, I think this is a great idea. I have sold other people on
> numpy/scipy, but myriad BLAS/LAPACK/fortran combinations makes the
> installation process unpleasant for new users.
same here :(
>> I have one issue:
>> scipy/sparse/linalg/dsolve/umfpack/umfpack.i:192: Error: Unable to
>> find 'umfpack.h'
>> I have installed libsuitesparse-dev providing /usr/include/suitesparse/umfpack.h
>> Of course, I can solve this with a symlink but it is very ugly.
>> It looks like there is something to fix in scipy auto-detection
>> because this "usr/include/suitesparse/umfpack.h" should be detected
>> and used.
> The "right" way to do this is to put a [umfpack] section in site.cfg
> Ideally, auto-detection would always include knowledge about where the
> major distributions put these files. Keep in mind that scipy's
> support for UMFPACK is deprecated in 0.7 and has been moved to a
Well I would like to have the feedback of the developers on the best
way to fix the auto-detection.
site.cfg should not be needed here.
>> Once again, my goal is to end up with something like:
>> 1) apt-get install "the correct list of packages"
>> 2) compile/install scipy svn *wthout any extra config*.
>> 3) you have a nice scipy installed and you can even check that using
>> scipy.test() (ok ok it is an svn version so if you get some errors
>> please report them ;))
> For 2) you can either improve the auto-detection (I don't know how
> myself) or simply not use UMFPACK. The danger is that someone might
> already have libsuitesparse-dev installed, so fixing the
> auto-detection is probably a better approach.
> Anyway, I'm eager to see this information become available. Do you
> plan to provide instructions for both 32-bit and 64-bit versions? In
> 8.04 I believe there were some differences in the availability of BLAS
> libraries between the two.
I do not need umfpack but my goal is to perform a scipy installation
with all the libs users could need.
I should be the same for 32 and 64 using 8.10 but I have to check that
(with kvm it should be easy)
My plan is to provide an installation procedure *without an hand
Ubuntu is going to release 8.10 end of october. It would be nice to have
a standard way to compile the svn on this "intrepid ibex".
Of course, the simplest way to having something working is apt-get
install python-scipy. The drawback with these nice packages it that you
cannot ask for patch or test a new feature.
If someone wants to write a procedure to compile a full optimized scipy
using ATLAS on 8.10 , he's really welcome :)
More information about the SciPy-user