[SciPy-User] Inconsistent scipy.test() failures

[Scott Sinclair]

On 20 June 2012 23:16, Russell Warren <russ at perspexis.com> wrote:
> I just tested what I thought was a successful scipy installation and it is
> failing.  What's more, it is failing differently between the first and
> subsequent runs.  FWIW, numpy.test("full") passes with no errors.
>
> Should I be concerned about any of these failures?  Are scipy test suite
> failures typical/expected?
>
> Results of first scipy.test() run (1 failure, peppered with warnings):
> http://pastebin.com/raw.php?i=00BrCYNA
>
> Results of subsequent scipy.test() runs (1 error, 3 failures, no warnings):
> http://pastebin.com/raw.php?i=DHcLUjp9
>
> System info:
>
> NumPy version 1.6.2
> SciPy version 0.10.1
> Python version 2.7.3 (default, Jun 19 2012, 16:53:07) [GCC 4.6.1]
> nose version 1.1.2
> lapack-3.4.1 (netlib version, compiled from
> http://www.netlib.org/lapack/lapack.tgz)
> gfortran v4.6.1
> gcc v4.6.1
> uname -a = Linux edms 3.0.21-tinycore #3021 SMP Sat Feb 18 11:54:11 EET 2012
> i686 GNU/Linux
> os.name: posix
> sys.platform: linux2

I don't think this is actually a problem. I only see this behaviour if
I run scipy.test() multiple times in the *same* Python interpreter
instance.

As I understand it, the warnings that are being tested for in
'test_mio.test_mat4_3d' and 'Regression test for #651: better handling
of badly conditioned' are only raised the first time they are
encountered in a given Python session - so those two tests will
obviously fail if you run scipy.test() multiple times.

I don't see the other failures, which look like minor numerical
precision problems (maybe due to cache effects?), I wouldn't worry too
much about those either.

Cheers,
Scott