[Numpy-discussion] Re: View on same data with different type

Florian Schulze florian.proff.schulze at gmx.net
Fri Dec 3 06:54:02 CST 2004


On Thu, 02 Dec 2004 21:49:10 +0100, Florian Schulze 
<florian.proff.schulze at gmx.net> wrote:

> Hi!
>
> I would like to be able to access the same array (memory location) with 
> arrays of different size and with different typecodes. Let's say I have 
> an array of 8 UInt8 and I want to view it as 2 UInt32. I want to be able 
> to change the content of either array and the change should be visible 
> in both arrays. To speak in C notation, I want an *UInt8 and a *Uint32 
> to the same memory location.
> Is that possible with Numeric or numarray, maybe even with slices of the 
> same data?
> The reason I want this is that I want to prevent copying memory around. 
> It would be even cooler if this would work with mmaped arrays, though 
> then it's enough when it would work with read only mmaps. BTW, why isn't 
> it allowed to create overlapping mmap slices?
>
> Regards,
> Florian Schulze

Hi again!

Todd Miller explained me how to do it (though it wasn't working as is, I 
figured it out). As I think it is interesting to other people, I post it 
here:

>>> import numarray
>>> a = numarray.arrayrange(12, type='i1')

# We can now create an array pointing to the same memory location with 
different type:

>>> b = numarray.array(sequence=a._data, shape=(3,), type='i4')

# The key here is to use a._data. You have to specify the shape, or else 
it doesn't work.

>>> a
array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11], type=Int8)
>>> b
array([ 50462976, 117835012, 185207048])

>>> a[3] = 10
>>> a
array([ 0,  1,  2, 10,  4,  5,  6,  7,  8,  9, 10, 11], type=Int8)
>>> b
array([167903488, 117835012, 185207048])

# Here you see how the assignment to a also changed b.

>>> a._data
<memory at 0x01031878 with size:0x0000000c held by object 0x00bcab40 
aliasing object 0x00000000>
>>> b._data
<memory at 0x01031878 with size:0x0000000c held by object 0x00bcab40 
aliasing object 0x00000000>

# Here you see that both arrays really point to the same memory location.


I tried it with numarray.records.array and it's basically the same, though 
you have to use 'buffer' instead of 'sequence'.

I will have to try it out with mmap, but Todd told me it should be no 
problem.

He also told me it could also be done with views by calling .view() on the 
array and then set the other values (_type, _shape, _itemsize, [_strides, 
_bytestride]) but I think this is more error prone and I didn't test it.

Thanks again to Todd Miller.

Regards,
Florian Schulze





More information about the Numpy-discussion mailing list