[NumPy-Tickets] [NumPy] #402: newaxis incompatible with array indexing

[NumPy Trac]

#402: newaxis incompatible with array indexing
---------------------------+------------------------------------------------
  Reporter:  NeilenMarais  |       Owner:  somebody   
      Type:  enhancement   |      Status:  closed     
  Priority:  normal        |   Milestone:  Unscheduled
 Component:  numpy.core    |     Version:             
Resolution:  wontfix       |    Keywords:             
---------------------------+------------------------------------------------

Comment(by mwiebe):

 Perhaps a better way to phrase what I mean is to remove the current
 interface to fancy indexing. It requires extra thought, and easily trips
 people up, as evidenced by a number of mailing list questions.

 {{{
 In [25]: a = np.arange(12).reshape(4,3)

 In [26]: a[[0,2,3]]
 Out[26]:
 array([[ 0,  1,  2],
        [ 6,  7,  8],
        [ 9, 10, 11]])

 # So lists in index 0 select rows

 In [27]: a[:,[2,1,2]]
 Out[27]:
 array([[ 2,  1,  2],
        [ 5,  4,  5],
        [ 8,  7,  8],
        [11, 10, 11]])

 # Cool, lists in index 1 select columns, let's put that together...

 In [28]: a[[0,2,3],[2,1,2]]
 Out[28]: array([ 2,  7, 11])

 # What happened? Totally unexpected, especially since the following works:

 In [31]: a[0:2,1:3]
 Out[31]:
 array([[1, 2],
        [4, 5]])

 # So when someone thinks "now let's replace the slice with a list because
 # I need rows/columns that aren't in a linear sequence," the behavior will
 # be confusing.

 }}}

 I would suggest the take function should be generalized to do multi-
 indexing as a replacement for the current fancy indexing.

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/402#comment:10>
NumPy <http://projects.scipy.org/numpy>
My example project