masked_array ** 2 operation differs from array ** 2 operation on cluster architecture

[andsor]

Hi @juliantaylor, it's me again. I encounter a similar issue as before (#6251) with the following code using now Python 3.4.3 (GCC 4.7.2) with numpy 1.9.3:

from __future__ import division, print_function, absolute_import

import numpy as np
import numpy.ma as ma

from pprint import pprint


a = ma.array(
    np.array(
        [0.25375743, 0.49557181, -0.32551902, -0.37675809, -0.04705212999999997]
#        [0.67696689, 0.91878127, 0.09769044, 0.04645137, 0.37615733]
    ),
    mask=np.array([False, False, False, False, False], dtype=np.bool)
)

pprint(a)

ma_squared = a**2
a = np.asarray(a)
a_squared = a**2

pprint(a_squared - ma_squared)
pprint(a_squared[0])
pprint(ma_squared[0])

The output is

masked_array(data = [0.25375743 0.49557181 -0.32551902 -0.37675809 -0.04705212999999997],
             mask = [False False False False False],
       fill_value = 1e+20)

masked_array(data = [ -1.38777878e-17   0.00000000e+00  -1.38777878e-17   0.00000000e+00
   0.00000000e+00],
             mask = False,
       fill_value = 1e+20)

0.064392833280204897
0.064392833280204911

numpy.show_config():

blas_opt_info:
    libraries = ['f77blas', 'cblas', 'atlas']
    language = c
    library_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/lib']
    include_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/include']
    define_macros = [('ATLAS_INFO', '"\\"3.8.4\\""')]
openblas_info:
  NOT AVAILABLE
lapack_opt_info:
    libraries = ['lapack', 'f77blas', 'cblas', 'atlas']
    language = f77
    library_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/lib']
    include_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/include']
    define_macros = [('ATLAS_INFO', '"\\"3.8.4\\""')]
atlas_3_10_blas_threads_info:
  NOT AVAILABLE
mkl_info:
  NOT AVAILABLE
atlas_blas_threads_info:
  NOT AVAILABLE
atlas_3_10_threads_info:
  NOT AVAILABLE
atlas_threads_info:
  NOT AVAILABLE
openblas_lapack_info:
  NOT AVAILABLE
lapack_mkl_info:
  NOT AVAILABLE
atlas_3_10_blas_info:
  NOT AVAILABLE
atlas_blas_info:
    libraries = ['f77blas', 'cblas', 'atlas']
    language = c
    library_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/lib']
    include_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/include']
    define_macros = [('ATLAS_INFO', '"\\"3.8.4\\""')]
atlas_info:
    libraries = ['lapack', 'f77blas', 'cblas', 'atlas']
    language = f77
    library_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/lib']
    include_dirs = ['/usr/nld/atlas-3.8.4-skadi-seq/include']
    define_macros = [('ATLAS_INFO', '"\\"3.8.4\\""')]
atlas_3_10_info:
  NOT AVAILABLE
blas_mkl_info:
  NOT AVAILABLE

The numpy tests pass, but this lets one scipy test fail, scipy/scipy#5197 .

Do you think it is related. This time I used the newer C compiler, though.

[juliantaylor]

I guess you are also using a really old libc?
the difference is likely due to masked array using pow(a, 2) while the non-masked rewrite this to a*a

[andsor]

Running the libc library:

 $ /lib64/libc.so.6
GNU C Library stable release version 2.11.1 (20100118), by Roland McGrath et al.
Copyright (C) 2009 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.
There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.
Configured for x86_64-suse-linux.
Compiled by GNU CC version 4.3.4 [gcc-4_3-branch revision 152973].
Compiled on a Linux 2.6.32 system on 2010-05-06.
Available extensions:
        crypt add-on version 2.1 by Michael Glad and others
        GNU Libidn by Simon Josefsson
        Native POSIX Threads Library by Ulrich Drepper et al
        BIND-8.2.3-T5B
For bug reporting instructions, please see:
<http://www.gnu.org/software/libc/bugs.html>.