[Numpy-discussion] pre-initialized arrays

[Stuart Brorson]

>> I realize NumPy != Matlab, but I'd wager that most users
>> would think that this is the natural behavior.
>
> I would not find it "natural" that elements of my float
> array could be assigned complex values.

OK, then NumPy should throw an exception if you try to make the 
assignemnt.

I tried it out.  NumPy does the right thing in this case:

In [10]: A = numpy.zeros([3, 3])

In [11]: A[1, 1] = 1+2j
---------------------------------------------------------------------------
<type 'exceptions.TypeError'>             Traceback (most recent call
last)

/fs/home/sdb/<ipython console> in <module>()

<type 'exceptions.TypeError'>: can't convert complex to float; use
abs(z)


However, not in this case:

In [12]: B = 10*numpy.random.rand(3, 3) + 1j*numpy.random.rand(3, 3)

In [13]: B[1, 1]
Out[13]: (7.15013107181+0.383220559014j)

In [14]: A[1, 1] = B[1, 1]

In [15]:


So the bug is that assignment doesn't do value checking in every
case.

>  How could it be
> a fixed chunk of memory and do such things, unless it would
> always waste enough memory to hold the biggest possible
> subsequent data type?

Fair question.  Matlab does a realloc if you assign complex values to
an initially real array.  It can take a long time if your matrices are
large.

Cheers,

Stuart Brorson
Interactive Supercomputing, inc.
135 Beaver Street | Waltham | MA | 02452 | USA
http://www.interactivesupercomputing.com/