# numpy: frequencies

Discussion in 'Python' started by robert, Nov 18, 2006.

1. ### robertGuest

I have an integer array with values limited to range(a,b) like:

ia=array([1,2,3,3,3,4,...2,0,1])

and want to speedly count the frequencies of the integers into get a density matrix.
Is this possible without looping?

Question 2: is it possible to compute a "moving maximum" without python looping

ia=array([4,2,1,5,3,2,2,0,1,1])
-> mvmax(ia,3) ->
[4,4,4,5,5,5,3,2,2,1])

Robert

robert, Nov 18, 2006

2. ### Filip WasilewskiGuest

robert wrote:
> I have an integer array with values limited to range(a,b) like:
>
> ia=array([1,2,3,3,3,4,...2,0,1])
>
> and want to speedly count the frequencies of the integers into get a density matrix.
> Is this possible without looping?

See numpy.bincount (for integers >= 0) if you mean 'without writing

> Question 2: is it possible to compute a "moving maximum" without python looping
>
> ia=array([4,2,1,5,3,2,2,0,1,1])
> -> mvmax(ia,3) ->
> [4,4,4,5,5,5,3,2,2,1])

I haven't seen a ready solution but this can be easily converted into
Pyrex/C looping.

cheers,
fw

Filip Wasilewski, Nov 18, 2006

3. ### Tim HochbergGuest

Filip Wasilewski wrote:
> robert wrote:
>> I have an integer array with values limited to range(a,b) like:
>>
>> ia=array([1,2,3,3,3,4,...2,0,1])
>>
>> and want to speedly count the frequencies of the integers into get a density matrix.
>> Is this possible without looping?

>
> See numpy.bincount (for integers >= 0) if you mean 'without writing
>
>> Question 2: is it possible to compute a "moving maximum" without python looping
>>
>> ia=array([4,2,1,5,3,2,2,0,1,1])
>> -> mvmax(ia,3) ->
>> [4,4,4,5,5,5,3,2,2,1])

>
> I haven't seen a ready solution but this can be easily converted into
> Pyrex/C looping.

I don't know a way to avoid looping entirely, but there are ways that
you can loop over the width of the window (in this case 3) rather than
over the entire array. Since the window width is generally small
compared to the array, this will probably be fast enough. The tricky
part is to get the value right at the edges, since what you do there
depends on what boundary conditions you apply.

The general idea is this:

result = ia[n-1:]
for i in range(n-1):
numpy.maximum(result, ia[i:-n+i], result)

This punts on dealing with the ends (and I haven't tested this version),
but should give you the idea.

-tim

Tim Hochberg, Nov 18, 2006