[Numpy-discussion] argsort

[Robert Kern]

On Wed, Jan 16, 2013 at 9:30 AM, Mads Ipsen <madsipsen at gmail.com> wrote:
> Hi,
>
> Thanks everybody for all the answers that make perfect sense when axis=0.
>
> Now suppose I want to sort the array in such a way that each row is sorted
> individually. Then I suppose I should do this:
>
> from numpy import *
>
>
> v = array([[4,3],
>            [1,12],
>            [23,7],
>            [11,6],
>            [8,9]])
> idx = argsort(v, axis=1)
>
> idx is then
>
> [[1 0]
>  [0 1]
>  [1 0]
>  [1 0]
>  [0 1]]
>
> which makes sense, since these are the indices in an order that would sort
> each row. But when I try
>
> a[idx, variuos_additional_arguments]
>
> I just get strange results. Anybody that can point me towards the correct
> solution.

Please have a look at the documentation again. If idx has indices for
the second axis, you need to put it into the second place.

  http://docs.scipy.org/doc/numpy/user/basics.indexing.html#indexing-multi-dimensional-arrays
  http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html#advanced-indexing


[~]
|4> idx0 = np.arange(v.shape[0])[:,np.newaxis]

[~]
|5> idx0
array([[0],
       [1],
       [2],
       [3],
       [4]])

[~]
|7> v[idx0, idx]
array([[ 3,  4],
       [ 1, 12],
       [ 7, 23],
       [ 6, 11],
       [ 8,  9]])

--
Robert Kern