Re: numpy errors in test-suite -- April 25, 2011 - 15:22

Hi,

As just discussed in a different post on the topic of the relax gui 
(https://mail.gna.org/public/relax-devel/2011-04/msg00052.html ; # 
Message-id: <4DB53313.7060505@xxxxxxxxx>), I manually switched to using 
32-bit python (2.6.6), numpy (1.5.1) and scipy (0.9.0) from fink 
(instead of 64-bit python 2.6.1, numpy 2.0.0, and scipy 0.10.0 directly 
under Mac OS X). As relax uses the "/usr/bin/env python" in order to get 
the python path (header of executable file "relax"), I had to manually 
change it to "/sw/bin/python2.6" (the fink path for python-2.6 binary) 
to do so.

The unit tests all passed without the warnings discussed before (post at 
https://mail.gna.org/public/relax-devel/2011-04/msg00050.html ; 
Message-id: <4DB419A4.1030201@xxxxxxxxx>). The warnings were also absent 
from the system tests, although 9 errors arose because of the 
unavailability of the C modules (discussed at 
https://mail.gna.org/public/relax-devel/2011-04/msg00052.html ; # 
Message-id: <4DB53313.7060505@xxxxxxxxx>).

Hence, it seems that these warnings arose because of either the 
numpy/scipy version, or the 32-bit/64-bit issue...

Is this worth fixing within relax's test-suite (as the tests passed, but 
with warnings) ???

All the best,


Séb  :)


On 11-04-24 2:37 PM, Sébastien Morin wrote:
> Hi,
>
> I just ran the numpy test in python with "numpy.test()" and it passed 
> with the following message:
>
>     OK (KNOWNFAIL=3, SKIP=1)
> <nose.result.TextTestResult run=3399 errors=0 failures=0>
>
> However, scipy (ran with "scipy.test()") got4 errors and 1 failure:
>
>     FAILED (KNOWNFAIL=12, SKIP=42, errors=4, failures=1)
> <nose.result.TextTestResult run=4766 errors=4 failures=1>
>
> The 4 errors were:
>
>     ERROR: test_complex_nonsymmetric_modes 
> (test_arpack.TestEigenComplexNonSymmetric)
>     ERROR: test_complex_symmetric_modes 
> (test_arpack.TestEigenComplexSymmetric)
>     ERROR: test_nonsymmetric_modes (test_arpack.TestEigenNonSymmetric)
>     ERROR: test_starting_vector (test_arpack.TestEigenNonSymmetric)
>
> and originated from the same problem:
>
>     ArpackError: ARPACK error -8: Error return from trid. eigenvalue 
> calculation; Informational error from LAPACK routine dsteqr.
>
> The failure was:
>
>     FAIL: test_expon (test_morestats.TestAnderson)
>
>
> Any idea if these problems cause the warning in relax's test-suite ?
>
> Thanks !
>
>
> Séb  :)
>
>
>
> On 11-04-24 2:00 PM, Sébastien Morin wrote:
> > Hi,
> >
> > When running relax's test-suite on Mac OS X, I get many times this
> > warning (although both the unit tests and system tests pass):
> >
> >       "Warning: invalid value encountered in double_scalars"
> >
> > This originated after my installation of numpy (version
> > 2.0.0.dev-ac2c160, running with scipy-0.10.0.dev) with relax-1.3 from
> > the repository (r12825).
> >
> > Any idea was causes these warnings and if they are important ?
> >
> > Thanks !
> >
> >
> > Séb

--
Sébastien Morin, Ph.D.
Postdoctoral fellow
S. Grzesiek NMR Laboratory
Biozentrum, Universität Basel
Basel, Switzerland