gmane.comp.python.scientific.devel

test_arpack errors on 64bit Linux

I am getting test_arpack errors on 64bit Linux, but not 32bit Linux.
Is anyone else seeing this?  I would like to get these fixed before
releasing the 0.7.0b1.

>>> scipy.test()
Running unit tests for scipy
NumPy version 1.2.0.dev5629
NumPy is installed in /usr/lib64/python2.5/site-packages/numpy
SciPy version 0.7.0.dev4637
SciPy is installed in /usr/lib64/python2.5/site-packages/scipy
Python version 2.5.1 (r251:54863, Jul 10 2008, 17:25:56) [GCC 4.1.2
20070925 (Red Hat 4.1.2-33)]
nose version 0.10.3

<snip>

======================================================================
ERROR: test_nonsymmetric_modes (test_arpack.TestEigenNonSymmetric)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib64/python2.5/site-packages/scipy/sparse/linalg/eigen/arpack/tests/test_arpack.py",
line 204, in test_nonsymmetric_modes
    self.eval_evec(m,typ,k,which)
  File "/usr/lib64/python2.5/site-packages/scipy/sparse/linalg/eigen/arpack/tests/test_arpack.py",
line 186, in eval_evec
    eval,evec=eigen(a,k,which=which,**kwds)
  File "/usr/lib64/python2.5/site-packages/scipy/sparse/linalg/eigen/arpack/arpack.py",
line 220, in eigen
    raise RuntimeError("Error info=%d in arpack"%info)
RuntimeError: Error info=-9 in arpack

(Continue reading)

headers

Aric Hagberg | 15 Aug 02:41

Re: test_arpack errors on 64bit Linux

On Thu, Aug 14, 2008 at 01:26:28PM -0700, Jarrod Millman wrote:
> I am getting test_arpack errors on 64bit Linux, but not 32bit Linux.
> Is anyone else seeing this?  I would like to get these fixed before
> releasing the 0.7.0b1.

I can verify that I get errors in the same tests - but slightly
different errors.  I see an info=-8 and info=-9999 return form ARPACK
in addition to info=-9 

c          = -8: Error return from LAPACK eigenvalue calculation;
c          = -9: Starting vector is zero.
c          = -10: IPARAM(7) must be 1,2,3,4.
c          = -11: IPARAM(7) = 1 and BMAT = 'G' are incompatable.
c          = -12: IPARAM(1) must be equal to 0 or 1.
c          = -9999: Could not build an Arnoldi factorization.
c                   IPARAM(5) returns the size of the current Arnoldi
c                   factorization.

I'm also seeing some other errors e.g

======================================================================
FAIL: test_dot (test_blas.TestFBLAS1Simple)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/u/aric/lib/python/scipy/lib/blas/tests/test_blas.py", line 69, in test_dot
    assert_almost_equal(f([3,-4,5],[2,5,1]),-9)
  File "/nh/u/aric/lib/python/numpy/testing/utils.py", line 207, in assert_almost_equal
    assert round(abs(desired - actual),decimal) == 0, msg
AssertionError: 
Items are not equal:

(Continue reading)

headers

Aric Hagberg | 16 Aug 16:51

Re: test_arpack errors on 64bit Linux

On Thu, Aug 14, 2008 at 06:41:37PM -0600, Aric Hagberg wrote:
> On Thu, Aug 14, 2008 at 01:26:28PM -0700, Jarrod Millman wrote:
> > I am getting test_arpack errors on 64bit Linux, but not 32bit Linux.
> > Is anyone else seeing this?  I would like to get these fixed before
> > releasing the 0.7.0b1.
> 
> I can verify that I get errors in the same tests - but slightly
> different errors.  I see an info=-8 and info=-9999 return form ARPACK
> in addition to info=-9 
> 
> Is there some LAPACK/BLAS issue?

It appears that the problem is with ATLAS package on 64bit Linux.

I rebuilt ATLAS from source (atlas-3.8.2) and all of the tests pass.  

This is the configuration that *doesn't* work (Ubuntu 64 bit):

>>> scipy.test()
Running unit tests for scipy
NumPy version 1.2.0.dev5654
NumPy is installed in /u/aric/lib/python/numpy
SciPy version 0.7.0.dev4645
SciPy is installed in /u/aric/lib/python/scipy
Python version 2.5.2 (r252:60911, Jul 31 2008, 17:31:22) [GCC 4.2.3
(Ubuntu 4.2.3-2ubuntu7)]
nose version 0.10.0
...
fail on arpack tests (and others)

(Continue reading)