Re: [Matplotlib-users] matplotlib:numarray slicing

[Robert K. <rk...@uc...>]

Graeme O'Keefe wrote:
> Hi,
> 
> I know this really belongs on the sourceforge.net:numarray list, but  it 
> is a very inactive list.

It's still the right place to ask. Your question would get answered in a 
timely manner. We're quiet; we're not gone.  :-)

Followup-to has been reset.

> I have noted the following with numarray:
> 
>  >>> x = numarray.zeros([3,3], numarray.Int16)
>  >>> i = [0,1,2,0,1,2]
>  >>> j = [0,1,2,0,1,2]
>  >>> x[i, j] += 1
>  >>> print x
> [[1, 0, 0]
>  [0, 1, 0]
> [0, 0, 1]]
> I was expecting (hoping)
> [[2, 0, 0]
>  [0, 2, 0]
>  [0, 0, 2]]
> 
> which is what you obviously get if you:
> 
> for n in range(len(i)) : x[i[n], j[n]] += 1
> 
> This is the sort of operation one does all the time when  histogramming 
> data-streams. Is there a way to achieve this without  looping over i/j.

This has been discussed recently on the numpy-discussion list (Subject: 
vectorizing histogram-like computation). The semantics of array indexing 
aren't anything like that for loop. I doubt that the two kinds of 
semantics could be merged consistently.

   x[i,j] += 1

is closer to

   x[i,j] = x[i,j] + 1

which amounts, in this case, to

   x[i,j] = array([1, 1, 1, 1, 1, 1])

There's nothing in this last operation to suggest that the numbers ought 
to be added together (and if the initial values weren't zero, a more 
complicated operation would have to considered).

In short, I consider it cleaner to keep specialized histogram code 
tucked away in a function which you can eventually optimize to C as 
required.

-- 
Robert Kern
rk...@uc...

"In the fields of hell where the grass grows high
  Are the graves of dreams allowed to die."
   -- Richard Harter