[Numpy-discussion] Strange bug in SVD when numpy is installed with pip

darcamo at gmail.com darcamo at gmail.com
Sat Sep 27 18:10:26 EDT 2014


Thanks Charles!

I installed atlas and lapack and then reinstalled numpy with pip. It works
correctly now.

An easy fix for this for now is raising an exception when numpy is using
its internal version for SVD and the user asked for the full matrices. The
exception message can instruct the user to either install lapack and
recompile numpy, or set full_matrices to False (if that is enough and the
user can't reinstall numpy with lapack for some reason). This would avoid
the bug hunting I had to do in my code.



Em 27/09/2014 17:37, "Charles R Harris" <charlesr.harris at gmail.com>
escreveu:

>
>
> On Sat, Sep 27, 2014 at 1:46 PM, Darlan Cavalcante Moreira <
> darcamo at gmail.com> wrote:
>
>>
>> >>> np.__config__.show()
>> lapack_info:
>>   NOT AVAILABLE
>> lapack_opt_info:
>>   NOT AVAILABLE
>> blas_info:
>>   NOT AVAILABLE
>> atlas_threads_info:
>>   NOT AVAILABLE
>> blas_src_info:
>>   NOT AVAILABLE
>> atlas_blas_info:
>>   NOT AVAILABLE
>> lapack_src_info:
>>   NOT AVAILABLE
>> openblas_info:
>>   NOT AVAILABLE
>> atlas_blas_threads_info:
>>   NOT AVAILABLE
>> blas_mkl_info:
>>   NOT AVAILABLE
>> blas_opt_info:
>>   NOT AVAILABLE
>> atlas_info:
>>   NOT AVAILABLE
>> lapack_mkl_info:
>>   NOT AVAILABLE
>> mkl_info:
>>   NOT AVAILABLE
>>
>>
>> With so many "NOT AVAILABLE" I'm now surprised SVD is even defined.
>>
>>
>>
> Looks like it is falling back to numpy's internal versions. You might try
> installing ATLAS to see if that makes a difference.
>
> BTW, the numpy custom is bottom posting.
>
> Chuck
>
>
> _______________________________________________
> NumPy-Discussion mailing list
> NumPy-Discussion at scipy.org
> http://mail.scipy.org/mailman/listinfo/numpy-discussion
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20140927/cd8be901/attachment.html>


More information about the NumPy-Discussion mailing list