[NumPy-Tickets] [NumPy] #1599: as_strided arrays are not reshape-able

[NumPy Trac]

#1599: as_strided arrays are not reshape-able
--------------------------------+-------------------------------------------
 Reporter:  dalleyg             |       Owner:  somebody
     Type:  defect              |      Status:  new     
 Priority:  normal              |   Milestone:  1.5.0   
Component:  numpy.core          |     Version:  1.4.0rc1
 Keywords:  as_strided reshape  |  
--------------------------------+-------------------------------------------
 Arrays created by as_strided are not reshape-able.  If one attempts to
 reshape, the operation will appear to succeed, but corrupted data will
 result.

 Here's a test example.
 {{{
 import numpy;
 print numpy.__version__;
 xx = numpy.arange(3, dtype=int)[None, None, :];
 xx2 = numpy.lib.stride_tricks.as_strided(xx, shape=(2,1,3),
 strides=xx.strides);
 xx3=xx2.reshape((2,3));
 print xx3
 }}}

 It prints things like
 {{{
 [[      0       1       2]
  [5602136      33       0]]
 }}}
 Note that the second row should look like the first, but it doesn't.

 This issue exists in numpy 1.3.0 and numpy 1.4.1 (at least).  I do not
 know if it persists in numpy 1.5.0.

 IMHO, reshape should: be generalized to handle  restrided arrays (at least
 ones that are restrided to perform a cheap tile operation), copy the array
 if the strides are too nonstandard, and/or raise an exception if the
 strides are too nonstandard.

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1599>
NumPy <http://projects.scipy.org/numpy>
My example project