Failed Unit test on macOSX about nnabla

Thanks for your feedback. We're currently working on Mac OS X support.

I was able to reproduce your error messages by following your steps to build NNabla using gcc (method No. 2). We're currently investigating further on this issue.

[yamachu]

My NumPy environment is below

>>> numpy.show_config()
lapack_opt_info:
    extra_link_args = ['-Wl,-framework', '-Wl,Accelerate']
    extra_compile_args = ['-msse3']
    define_macros = [('NO_ATLAS_INFO', 3), ('HAVE_CBLAS', None)]
openblas_lapack_info:
  NOT AVAILABLE
atlas_3_10_blas_threads_info:
  NOT AVAILABLE
atlas_threads_info:
  NOT AVAILABLE
atlas_3_10_threads_info:
  NOT AVAILABLE
atlas_blas_info:
  NOT AVAILABLE
atlas_3_10_blas_info:
  NOT AVAILABLE
atlas_blas_threads_info:
  NOT AVAILABLE
openblas_info:
  NOT AVAILABLE
blas_mkl_info:
  NOT AVAILABLE
blas_opt_info:
    extra_link_args = ['-Wl,-framework', '-Wl,Accelerate']
    extra_compile_args = ['-msse3', '-I/System/Library/Frameworks/vecLib.framework/Headers']
    define_macros = [('NO_ATLAS_INFO', 3), ('HAVE_CBLAS', None)]
blis_info:
  NOT AVAILABLE
atlas_info:
  NOT AVAILABLE
atlas_3_10_info:
  NOT AVAILABLE
lapack_mkl_info:
  NOT AVAILABLE

See blas_opt_info, use vecLib (maxOSX vector default library).
I think it would be nice to change using not vecLib, but OpenBLAS.

I'll check it.

Re-install NumPy with using OpenBLAS, but result was same...

[TakuyaNarihira]

Thanks for your investigation. In general, reduction operation is sensitive to its algorithms in terms of numerical errors. The difference possibly comes from 1) the backends of Numpy, or 2) the possible difference of behavior of Eigen library between OSX and other OSs. We use Eigen in sum in CPU implementation.