[Numpy-discussion] Problems casting object arrays to string type on Ubuntu

Dan Yamins dyamins@gmail....
Tue Nov 24 08:43:21 CST 2009


Really, no idea about this?   (Sorry if my original email was unclear.)

On Sat, Nov 21, 2009 at 3:27 PM, Dan Yamins <dyamins@gmail.com> wrote:

> Hi all,
>
> I'm having some issues casting object arrays to string type, especially on
> my Ubuntu installation. (Ubuntu Jaunty, 9.04, with Numpy v. 1.3.)
>
> With small arrays, the conversion is just wrong.  With large arrays, there
> seems to be some memory corruption.   Conversion to int or float (when
> appropriate) seems to work.
>
> For instance, here are some examples of errors with small arrays:
>
>   >>> X = numpy.array([13,14],'object')
>   >>> X.astype('|S2')
>   array(['13', '\xb2'],
>       dtype='|S2')
>   >>> X.astype('int')      #conversion to int or float seems to work fine
>   array([13, 14])
>   >>> X.astype(float)
>   array([ 13.,  14.])
>   >>> X = numpy.array(['cat','bat'],'object')
>   >>> X.astype('|S3')
>   array(['cat', '\x00ba'],
>        dtype='|S3')
>
> Large arrays:
>
>    In [24]: X = numpy.array(['cat','bat']*300000,'object')
>    In [25]: Y = X.astype('|S3')
>    *** glibc detected *** /usr/bin/python: munmap_chunk(): invalid pointer:
> 0xb7cd5008 ***
>
>
> I do NOT have this problem with Numpy 1.4.0.dev7746, installed on my OSX
> 10.5.8 machine.   There, everything seems to work fine.
>
> What's going on?  I feel like I've seem some traffic about related issues
> on the list before, but I couldn't quite tell from reading the threads what
> the "final upshot" of the discussion was ...    Is this something that was
> fixed in recent NumPy 1.4 releases, or is something about my Ubuntu vs. OSX
> installations?    Generally speaking can I / should I be relying on casting
> from object arrays to do the "right" (or "intuitive") thing?
>
> thanks,
> Dan
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/numpy-discussion/attachments/20091124/e8b6eae5/attachment.html 


More information about the NumPy-Discussion mailing list