[PYTHON MATRIX-SIG] take() in 1.0a2 (longish, sorry)

[Carlos Fonseca]

Hi,

I don't know if the following is intentional or not:

>>> from Numeric import *
>>> a=array([[2,3,7,6,4,7],[6,9,6,4,3,2]])
>>> a
2 3 7 6 4 7
6 9 6 4 3 2
>>> sort(a,-1)
2 3 4 6 7 7
2 3 4 6 6 9
>>> argsort(a,-1)
0 1 4 3 2 5
5 4 3 0 2 1
>>> take(a,argsort(a,-1),-1)
2 3 4 6 7 7
7 4 6 2 7 3

6 9 3 4 6 2
2 3 4 6 6 9
>>> 

I.e., when take is given a 2 dimensional indices array, it adds yet
another dimension to a. This should not happen if take(a,argsort(a,-1),-1) 
is to be equivalent to sort(a,-1). Note that, in the example above,
indices is perfectly aligned with a.

In general, if 

        b=take(a,indices,axis)

then

        b[i,...,j,...,k]=a[i,...,indices[i,...,j,...,k],...,k]
                ^
              axis 

with broadcasting of dimensions for all axes other than the "active" axis. 

Since a and indices would normally be aligned at the last dimension,
at least the following should be true:

a
2 3 7 6 4 7
6 9 5 4 3 2

take(a,[1,0,1],-1)  # select columns
3 2 3
9 6 9

take(a,[1,0,1],-2)
Arrays not aligned!

take(a,[[1],[0],[1]],-1)
Arrays not aligned!

take(a,[[1],[0],[1]],-2) # select rows
6 9 5 4 3 2
2 3 7 6 4 7
6 9 5 4 3 2

As for the default axes, oh well... :-)

Carlos



=================
MATRIX-SIG  - SIG on Matrix Math for Python

send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
=================