[Numpy-discussion] Expected behavior of numpy object arrays. Is this a bug?

[Robert Kern]

Rudolph van der Merwe wrote:
> Can someone please confirm if the following is the expected behavior
> of numpy ndarrays of dtype=object, i.e. object arrays. I suspect it
> might be a bug.

It's expected. The array() function has to make some guesses as to what you
meant when you pass it a sequence of sequences and are trying to make an object
array. It tries to go as deep as it can. When you give it a list of len-2
sequences, it thinks you want a (2, 2) array. When you give it a list of a len-2
sequence an a len-3 sequence, the only thing it can do is make a (2,) array of
the two objects.

The best way to build the array you want is to make the object array first, and
assign the contents:


In [1]: from numpy import *

In [2]: a = array([1, 2])

In [3]: b = array([5, 6, 7])

In [4]: c = array([3, 4])

In [5]: oa1 = empty([2], dtype=object)

In [6]: oa2 = empty([2], dtype=object)

In [7]: oa1[:] = [a, b]

In [8]: oa2[:] = [a, c]

In [9]: oa1
Out[9]: array([[1 2], [5 6 7]], dtype=object)

In [10]: oa2
Out[10]: array([[1 2], [3 4]], dtype=object)

In [11]: oa1[0].dtype
Out[11]: dtype('int32')

In [12]: oa2[0].dtype
Out[12]: dtype('int32')

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
 that is made terrible by our own mad attempt to interpret it as though it had
 an underlying truth."
  -- Umberto Eco