[NumPy-Tickets] [NumPy] #1870: Cannot assign a list-of-lists to dtype=object ndarray if sublists are equal-length

[NumPy Trac]

#1870: Cannot assign a list-of-lists to dtype=object ndarray if sublists are
equal-length
-------------------------+--------------------------------------------------
  Reporter:  mjuric      |       Owner:  somebody
      Type:  defect      |      Status:  reopened
  Priority:  high        |   Milestone:  1.6.2   
 Component:  numpy.core  |     Version:  1.6.0   
Resolution:              |    Keywords:          
-------------------------+--------------------------------------------------

Comment(by mwiebe):

 I believe the new example is invalid, but the error message ended up being
 of poorer quality than would be ideal because of an important backwards-
 compatibility decision:

 http://mail.scipy.org/pipermail/numpy-discussion/2011-March/055318.html

 This should illustrate the incompatibility of the assignment being done:

 {{{
 >>> a = numpy.zeros((5,5))
 >>> v = numpy.ones((5,1))
 >>> a[:,3]
 array([ 0.,  0.,  0.,  0.,  0.])
 >>> v
 array([[ 1.],
        [ 1.],
        [ 1.],
        [ 1.],
        [ 1.]])
 >>> a[:,3].shape
 (5,)
 >>> v.shape
 (5, 1)
 }}}

 The (5,) and (5,1) get broadcast to a (5,5) shape, which is what is really
 then being assigned into the (5,) array. To fix it, write the assignment
 like a[:,3] = v[:,0].

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1870#comment:8>
NumPy <http://projects.scipy.org/numpy>
My example project