Thanks mark! I think that's exactly what I'm looking for. We even had a previous discussion about this (oops!) (http://mail.scipy.org/pipermail/numpy-discussion/2011-January/054421.html).

I didn't find any documentation, I will try to add some once I understand how it works better.

John

On Sat, Oct 1, 2011 at 2:53 PM, Mark Wiebe <mwwiebe@gmail.com> wrote:
On Sat, Oct 1, 2011 at 1:45 PM, John Salvatier <jsalvati@u.washington.edu> wrote:
I apologize, I picked a poor example of what I want to do. Your suggestion would work for the example I provided, but not for a more complex example. My actual task is something like a "group by" operation along a particular axis (with a known number of groups). 

Let me try again: What I would like to be able to do is to specify some of the iterator dimensions to be handled manually by me. For example lets say I have some kind of a 2d smoothing algorithm. If I start with an array of shape [a,b,c,d] and I'd like to do the 2d smoothing over the 2nd and 3rd dimensions, I'd like to be able to tell nditer to do normal broadcasting and iteration over the 1st and 4th dimensions but leave iteration over the 2nd and 3rd dimensions to me and my algorithm. Each iteration of nditer would give me a 2d array to which I apply my algorithm. This way I could write more arbitrary functions that operate on arrays and support broadcasting.

Is clearer?

Maybe this will work for you:

In [15]: a = np.arange(2*3*4*5).reshape(2,3,4,5)

In [16]: it0, it1 = np.nested_iters(a, [[0,3], [1,2]], flags=['multi_index'])

In [17]: for x in it0:
   ....:     print it1.itviews[0]
   ....:     
[[ 0  5 10 15]
 [20 25 30 35]
 [40 45 50 55]]
[[ 1  6 11 16]
 [21 26 31 36]
 [41 46 51 56]]
[[ 2  7 12 17]
 [22 27 32 37]
 [42 47 52 57]]
[[ 3  8 13 18]
 [23 28 33 38]
 [43 48 53 58]]
[[ 4  9 14 19]
 [24 29 34 39]
 [44 49 54 59]]
[[ 60  65  70  75]
 [ 80  85  90  95]
 [100 105 110 115]]
[[ 61  66  71  76]
 [ 81  86  91  96]
 [101 106 111 116]]
[[ 62  67  72  77]
 [ 82  87  92  97]
 [102 107 112 117]]
[[ 63  68  73  78]
 [ 83  88  93  98]
 [103 108 113 118]]
[[ 64  69  74  79]
 [ 84  89  94  99]
 [104 109 114 119]]

Cheers,
Mark


 


On Fri, Sep 30, 2011 at 5:04 PM, Mark Wiebe <mwwiebe@gmail.com> wrote:
On Fri, Sep 30, 2011 at 8:03 AM, John Salvatier <jsalvati@u.washington.edu> wrote:
Using nditer, is it possible to manually handle dimensions  with different lengths? 

For example, lets say I had an array A[5, 100] and I wanted to sample every 10 along the second axis so I would end up with an array B[5,10]. Is it possible to do this with nditer, handling the iteration over the second axis manually of course (probably in cython)?


@cython.boundscheck(False)
def sum_squares_cy(arr):
    cdef np.ndarray[double] x
    cdef np.ndarray[double] y
    cdef int size
    cdef double value
    cdef int j

    axeslist = list(arr.shape)
    axeslist[1] = -1

    out = zeros((arr.shape[0], 10))
    it = np.nditer([arr, out], flags=['reduce_ok', 'external_loop',
                                      'buffered', 'delay_bufalloc'],
                op_flags=[['readonly'], ['readwrite', 'no_broadcast']],
                op_axes=[None, axeslist],
                op_dtypes=['float64', 'float64'])
    it.operands[1][...] = 0
    it.reset()
    for xarr, yarr in it:
        x = xarr
        y = yarr
        size = x.shape[0]
        j = 0
        for i in range(size):
           #some magic here involving indexing into x[i] and y[j]
    return it.operands[1]

Does this make sense? Is it possible to do?

I'm not sure I understand precisely what you're asking. Maybe you could reshape A to have shape [5, 10, 10], so that one of those 10's can match up with the 10 in B, perhaps with the op_axes?

-Mark
 

_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion



_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion



_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion



_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion