[SciPy-user] Re: SciPy installation

eric eric at scipy.org
Thu Apr 11 08:54:02 CDT 2002


Hey Mark,

Your best shot at getting SciPy installed is actually to get the version in CVS.
By now, it is actually more stable than the 0.1 release, easier to install, and
has a lot more features.  A little known fact is that a tarball of the latest
snapshot (last night) of the CVS is at:

    ftp://www.scipy.org/pub/scipy_snapshot.tgz

If your not comfortable with CVS, grab this.  The setup.py file should work
fine, but there are a few dependencies:

    f2py2e, ATLAS, and Numeric.

Note that fftw is no longer needed -- fftpack comes with SciPy now.

f2py2e:

f2py is easy to get and install.  Be sure to get a recent copy:

    http://cens.ioc.ee/projects/f2py2e/2.x/F2PY-2.13.175-1250.tar.gz

f2py comes with a setup file, so just unpack this, cd to the directory and run:

    python2 setup.py install --prefix=~

This will install the python files in:

    ~/lib/Python2.x/site-packages/

You'll need to add the given path to your PYTHONPATH.

ATLAS:

SciPy relies on a developer release of ATLAS which is ATLAS 3.3.14.  Its a bit
of a pain to build (takes a while and has a couple of steps for full lapack
support), so I've put a copy of the one I use here:

   http://www.scipy.org/Members/eric/atlas-RH7.1-PIII.tgz/

If your on a pre-PIII mahcine, then it probably won't work.  If your not using
RH7.1, your milage may also vary.

Unpack this file in ~/lib.  It should create ~/lib/atlas.

If you have to build it yourself, it takes anywhere from 15 minutes to 4 hours
to build.  After you finish, you need to follow the instructions here to get a
fully
built lapack library:

    http://math-atlas.sourceforge.net/errata.html#completelp

Hmmm.  I guess your also gonna need to build the "slow" lapack to get this to
work.

    http://www.netlib.org/lapack/

Numeric:

Grab the Numeric-21.0 version from:

  http://prdownloads.sourceforge.net/numpy/Numeric-21.0.tar.gz

Install with:

    python2 setup.py install --prefix=<your local python directory>

SciPy:
Once f2py2e, ATLAS, and Numeric are installed and you've unpacked the scipy
tarball, your almost set.  Because your ATLAS libraries are not installed in
/usr/local, you'll probably need to copy the
scipy/scipy_distutils/sample_site.cfg to scipy/scipy_distutils/site.cfg and edit
it to tell scipy where to find atlas.  Then, run the following from the scipy
directory:

    python2 setup.py install --prefix=<your local python directory>

If all goes well, you've ready to use scipy.  Also, please let me know if find
that any of these directions need correcting.

If you have more questions, please post to scipy-user at scipy.org or
scipy-dev at scipy.org.  There are a lot more people there, and you'll get answers
more quickly and perhaps targeted to your platform a little better.

good luck
eric


----- Original Message -----
From: "Mark Fardal" <fardal at coral.phys.uvic.ca>
To: <eric at enthought.com>
Sent: Wednesday, April 10, 2002 11:43 PM
Subject: SciPy installation


>
> Hi,
>
> I already wrote about my failure to install from RPM on Red Hat
> 7.1/7.2.  I had unpacked a SciPy-0.1 directory and noticed the
> setup.py procedure there.  Thought I'd give it a go...I can't tell how
> far this is from working, perhaps you would know.
>
> I changed python to python2 (damn Red Hat) in setup.py.
> Ran, tons of warning messages, failed on fftw
>
> gcc -shared
build/temp.linux-i686-2.1/fftw_wrap.o -L/astro/fardal/lib -L/usr/local/lib/gcc-l
ib/i686-pc-linux-gnu/2.95.3/specs -L/usr/local/lib/gcc-lib/i686-pc-linux-gnu/2.9
5.3/specs -Lbuild/temp.linux-i686-2.1 -Lbuild/temp.linux-i686-2.1 -lfftw_threads
 -lrfftw_threads -lfftw -lrfftw -lpthread -lamos -ltoms -lfitpack -lminpack -lqu
adpack -lodepack -llinpack_lite -lblas -lmach -lg2c -lgist -lc_misc -lcephes -o
build/lib.linux-i686-2.1/scipy/fft/fftw.so
> /usr/bin/ld: cannot find -lfftw_threads
> collect2: ld returned 1 exit status
> error: command 'gcc' failed with exit status 1
>
> Tried again, setting build_fftw=0
> still lots of warnings, but no failures
>
> warnings are things like:
> warning: build_py: package init file './compiler/tests/__init__.py' not found
(or not a regular file)
> special/cephes/ellie.c:58: warning: function declaration isn't a prototype
> interpolate/fitpack/fppogr.f:16: warning: `ve' might be used uninitialized in
this function
> Numerical/Src/dlapack_lite.c:9579: warning: suggest parentheses around &&
within ||
> etc...I'm assuming they are innocuous
>
> Don't want to do setup.py install since it'll probably go wrong place...
> I have no access to /usr/local, which is the problem with using the rpm.
> pelican:SciPy-0.1[ 47 ] ls build/lib.linux-i686-2.1/
> gui_thread  Numeric  scipy
>
> I just moved these directories to my comp/python/lib dir which is in my
> PYTHONPATH
>
> Try to use: wants fftw
> eliminated references to fftw and handy
>
> Complains about no fastumath
>
> Where's fastumath?  in the new Numeric directory, fastumath.so
> ah! I have python1.5/site-packages/Numeric in my PYTHONPATH too.  eliminate
> this...oops, no dice:
>
> python> import Numeric
> Traceback (most recent call last):
>   File "<stdin>", line 1, in ?
> ImportError: No module named Numeric
>
> There is no __init__.py in the new Numeric.  As best as I can tell
> (newbie alert), it needs one to count as a module.  Why didn't I get
> an __init__.py?  If it's at all educational, this is what my directories
> look like now.
>
> pelican:scipy[ 107 ] ls
> build_clib_new.py   ga            interpolate      scipy_tempfile.py
> build_flib.py       gplt          io               scipy_test.py
> cluster             handy.py      limits.py        scipy_version.py
> compiler            handy.pyc     misc.py          scipy_version.pyc
> cow                 helpmod.py    misc.py~         signal
> data_store.py       helpmod.pyc   misc.pyc         special
> dumbdbm_patched.py  __init__.py   optimize         stats
> dumb_shelve.py      __init__.py~  plt              sync.py
> fft                 __init__.pyc  proc.py          tests
> fortran_support.py  integrate     saved_values.py  xplt
>
> pelican:Numeric[ 105 ] ls
> arrayfns.so       LinearAlgebra.py  Numeric.pyc          Precision.pyc
> ArrayPrinter.py   Matrix.py         numeric_version.py   RandomArray.py
> ArrayPrinter.pyc  MLab.py           numeric_version.pyc  ranlib.so
> fastumath.so      multiarray.so     _numpy.so            umath.so
> lapack_lite.so    Numeric.py        Precision.py         UserArray.py
>
> any advice?
>
> thanks,
> Mark Fardal
> University of Victoria
>





More information about the SciPy-user mailing list