[Python-modules-team] Bug#581043: python-numpy: veiled problem interfacing lapack routines while providing ndarray of non-native byte_order

[Yaroslav Halchenko]

Package: python-numpy
Version: 1:1.3.0-3+b1
Severity: important
Tags: upstream


While working on resolving FTBFS #580879 against scikit-learn, I've ran into
numpy being unable to compute properly some (not sure if all or exhaustive list
of what are affected) functions from LAPACK interfaced under .linalg, whenever
ndarray's dtype is of non-native byteorder.

for instance on my amd64 laptop:

$> python -c "import numpy as N; print N.linalg.cholesky(N.array([[ 5.7998084,  -2.1825367 ], [-2.1825367,   9.85910595]], dtype='<f8'))"
[[ 2.40827914 -0.        ]
 [-0.90626401  3.00629199]]
                          
$> python -c "import numpy as N; print N.linalg.cholesky(N.array([[ 5.7998084,  -2.1825367 ], [-2.1825367,   9.85910595]], dtype='>f8'))"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/lib/python2.5/site-packages/numpy/linalg/linalg.py", line 423, in cholesky
    Cholesky decomposition cannot be computed'
numpy.linalg.linalg.LinAlgError: Matrix is not positive definite -         Cholesky decomposition cannot be computed
                          
$> python -c "import numpy as N; print N.linalg.svd(N.array([[ 5.7998084,  -2.1825367 ], [-2.1825367,   9.85910595]], dtype='<f8'))"
(array([[-0.39937903,  0.9167859 ],
       [ 0.9167859 ,  0.39937903]]), array([ 10.80988342,   4.84903093]), array([[-0.39937903,  0.9167859 ],
       [ 0.9167859 ,  0.39937903]]))
                          
$> python -c "import numpy as N; print N.linalg.svd(N.array([[ 5.7998084,  -2.1825367 ], [-2.1825367,   9.85910595]], dtype='>f8'))"
(array([[ 0., -1.],                                                        
       [-1.,  0.]]), array([  1.29365302e+157,   1.29365302e+157]), array([[ 1.,  0.],
       [ 0.,  1.]]))

so, some times it might fail with incorrect error msg for the exception, some
times just produces bogus results.

-- System Information:
Debian Release: squeeze/sid
  APT prefers unstable
  APT policy: (901, 'unstable'), (900, 'testing'), (300, 'experimental')
Architecture: amd64 (x86_64)

Kernel: Linux 2.6.32-3-amd64 (SMP w/2 CPU cores)
Locale: LANG=en_US, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash

Versions of packages python-numpy depends on:
ii  libatlas3gf-base [liblapack.s 3.8.3-19   Automatically Tuned Linear Algebra
ii  libblas3gf [libblas.so.3gf]   1.2-7      Basic Linear Algebra Reference imp
ii  libc6                         2.10.2-7   Embedded GNU C Library: Shared lib
ii  libgcc1                       1:4.4.3-9  GCC support library
ii  libgfortran3                  4.4.3-9    Runtime library for GNU Fortran ap
ii  liblapack3gf [liblapack.so.3g 3.2.1-7    library of linear algebra routines
ii  python                        2.5.4-9    An interactive high-level object-o
ii  python-central                0.6.16     register and build utility for Pyt

python-numpy recommends no packages.

Versions of packages python-numpy suggests:
ii  python-nose                 0.11.1-1     test discovery and running for Pyt
ii  python-numpy-dbg            1:1.3.0-3+b1 Fast array facility to the Python 
ii  python-numpy-doc            1:1.3.0-3    NumPy documentation

-- no debconf information