[Numpy-discussion] integer array creation oddity

Charles R Harris charlesr.harris@gmail....
Mon Jul 21 17:25:56 CDT 2008

On Mon, Jul 21, 2008 at 3:37 PM, Stéfan van der Walt <stefan@sun.ac.za>

> 2008/7/21 Suchindra Sandhu <suchindra@gmail.com>:
> > Is that the recommended way of checking the type of the array? Ususally
> for
> > type checkin, I use the isinstance built-in in python, but I see that
> will
> > not work in this case. I must admit that I am a little confused by this.
> Why
> > is type different from dtype?
> Data-types contain additional information needed to lay out numerical
> types in memory, such as byte-order and bit-width.  Each data-type has
> an associated Python type, which tells you the type of scalars in an
> array of that dtype.  For example, here are two NumPy data-types that
> are not equal:
> In [6]: d1 = np.dtype(int).newbyteorder('>')
> In [7]: d2 = np.dtype(int).newbyteorder('<')
> In [8]: d1.type
> Out[8]: <type 'numpy.int32'>
> In [9]: d2.type
> Out[9]: <type 'numpy.int32'>
> In [10]: d1 == d2
> Out[10]: False
> I don't know why there is more than one int32 type (I would guess it
> has something to do with the way types are detected upon build; maybe
> Robert or Travis could tell you more).

They correspond to two C types of the same size, int and long. On 64 bit
systems you should have two int64 types, long and longlong.

In [1]: dtype('i').name
Out[1]: 'int32'

In [2]: dtype('l').name
Out[2]: 'int32'

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/numpy-discussion/attachments/20080721/21293122/attachment.html 

More information about the Numpy-discussion mailing list