[SciPy-Dev] spatial.test() failure

[Warren Weckesser]

>
> Nils Wagner wrote:
> >>>> from scipy import spatial
> >>>> spatial.test()
> >>>>
> > Running unit tests for scipy.spatial
> > NumPy version 2.0.0.dev8714
> > NumPy is installed in
> > /home/nwagner/local/lib64/python2.6/site-packages/numpy
> > SciPy version 0.9.0.dev6803
> > SciPy is installed in
> > /home/nwagner/local/lib64/python2.6/site-packages/scipy
> > Python version 2.6.5 (r265:79063, Jul  5 2010, 11:46:13)
> > [GCC 4.5.0 20100604 [gcc-4_5-branch revision 160292]]
> > nose version 0.11.2
> > 
> ............................................................................................................................F..................................................................................................................................................
> > ======================================================================
> > FAIL: Tests pdist(X, 'minkowski') on iris data. (float32)
> > ----------------------------------------------------------------------
> > Traceback (most recent call last):
> >    File
> > 
> "/home/nwagner/local/lib64/python2.6/site-packages/scipy/spatial/tests/test_distance.py",
> > line 837, in test_pdist_minkowski_3_2_iris_float32
> >      self.failUnless(within_tol(Y_test1, Y_right, eps))
> > AssertionError
> >
> > ----------------------------------------------------------------------
> > Ran 271 tests in 38.005s
> >
> > FAILED (failures=1)
> > <nose.result.TextTestResult run=271 errors=0 failures=1>
> >
>
>
> In r6799, the names of some tests were changed so there are no longer
> duplicated test function names.  This means tests are now being run that
> had previously been shadowed by the duplicate names.  I also get the
> failure of the test test_pdist_minkowski_3_2_iris_float32().  I created
> a ticket for the problem: http://projects.scipy.org/scipy/ticket/1278
>
> Warren
>

If the tolerance in the test is changed from eps=1e-7 to 1e-6, the test 
passes on my computer.  Several other float32 tests using the iris data 
also use 1e-6.  I added a note about the tolerance of similar tests in 
the ticket.  It does not seem unreasonable to simply loosen the 
tolerance a bit. Any objections?

Warren

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20100926/68959b5f/attachment.html>