[SciPy-user] Different failures in scipy.test()

Pearu Peterson pearu at scipy.org
Wed Oct 1 15:30:41 EDT 2003


On Wed, 1 Oct 2003, Nils Wagner wrote:

> >> scipy.__version__
> '0.2.0__220.4275'
> >>> Numeric.__version__
> '23.0'
> Python 2.2.2 (#1, Mar 17 2003, 15:17:58)
> [GCC 3.3 20030226 (prerelease) (SuSE Linux)] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
> 
> 
> BTW, Is it possible to dtermine the version of ATLAS in a similar way ?

Yes, see INSTALL.txt: section TROUBLESHOOTING, item (6)

> ======================================================================
> FAIL: check_pro_ang1_cv (test_basic.test_cephes)
> ----------------------------------------------------------------------
> Items are not equal:
> DESIRED: (1.0, 0.0)
> ACTUAL: (0.99999999999999989, 0.0)

This could be related to Python 2.2.2, not sure. On Python 2.3.1 this 
test succeeds.

> #################################################################################################################
> 
> The same cvs version of scipy
> but
> >>> Numeric.__version__
> '23.1'

> ======================================================================
> FAIL: check_sh_chebyt (test_basic.test_sh_chebyt)
> ----------------------------------------------------------------------
> ======================================================================
> FAIL: check_random_complex (test_decomp.test_cholesky)
> ----------------------------------------------------------------------
> ======================================================================
> FAIL: check_simple_complex (test_decomp.test_cholesky)
> ----------------------------------------------------------------------

I am not sure why these tests fail when switching Numeric 23.0->23.1.
I am using Python 2.3.1, Numeric 23.1, Atlas 3.2.1 and Scipy 
0.2.0__220.4275 and all tests pass OK. I'll later try earlier versions of 
Python...

Anyway, I don't see how the above failures could be related to
using different ATLAS versions as you originally suggested.

Pearu



More information about the SciPy-User mailing list