[NumPy-Tickets] [NumPy] #2069: boolean indexing with no matches should leave shape same

[NumPy Trac]

#2069: boolean indexing with no matches should leave shape same
--------------------+-------------------------------------------------------
 Reporter:  shaunc  |       Owner:  somebody   
     Type:  defect  |      Status:  new        
 Priority:  normal  |   Milestone:  Unscheduled
Component:  Other   |     Version:  1.6.1      
 Keywords:          |  
--------------------+-------------------------------------------------------

Old description:

> >>> a = array( [ [ 1, 1 ] ] )
> >>> a[ a == 0 ].shape
> (0, )
>
> This is also puzzling as:
> >>> aa == 0
> array([[False, False]], dtype=bool)
>
> >>> b = [ [ False, False ] ]
> >>> a [ b ]
> >>> aa[ bb ]
> array([[1, 1],
>         [1, 1]])

New description:

 {{{
 >>> a = array( [ [ 1, 1 ] ] )
 >>> a[ a == 0 ].shape
 (0, )
 }}}

 This is also puzzling as:
 {{{
 >>> aa == 0
 array([[False, False]], dtype=bool)

 >>> b = [ [ False, False ] ]
 >>> a [ b ]
 >>> aa[ bb ]
 array([[1, 1],
         [1, 1]])
 }}}

--

Comment(by mwiebe):

 The first thing you're seeing is by design. NumPy picks out the the array
 elements which are flagged with True, and returns them compressed into a
 one-dimensional array.

 The second thing you're seeing is a bug. The fancy indexing code is
 treating the [[False, False]] as an integer array, so is using the False
 values as an integer index instead of as a boolean. If you force it to be
 a boolean array, it works as expected:

 {{{
 In [10]: a[b]
 Out[10]:
 array([[1, 1],
        [1, 1]])

 In [11]: a[np.array(b)]
 Out[11]: array([], dtype=int32)
 }}}

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/2069#comment:1>
NumPy <http://projects.scipy.org/numpy>
My example project