Re: iterating cell arrays

[David Bateman]

According to John W. Eaton <address@hidden> (on 06/25/04):
> On 25-Jun-2004, David Bateman <address@hidden> wrote:
> 
> | Ok, so if x is an NDArray or ND Cell array, Matlab iterates 
> | prod(size(a)(2:end)), passing column vectors of length size(a,1). 
> | This doesn't make any sense to me, since why is the first dimension
> | privileged in this manner. To be consistent with the use of
> | NDArrays elsewhere, the for loop should accept a "dim" argument that
> | specifies the dimension over which the array is iterated. But that
> | is likely to be very ugly. 
> 
> I think this feature is a holdover from the early days of Matlab, when
> 
>   for i = X
>     ...
>   end
> 
> assigned each column of X in turn to i for each trip through the
> loop.  Since a colon expression like 1:N is really a matrix object
> with one row, it is completely consistent to have
> 
>   for i = 1:N
> 
> assign 1, 2, ..., N to i in turn for each trip through the loop.
> 
> Once N-d arrays were introduced, I suppose it made some sense to
> continue iterating over the all the "columns" of X, starting with the
> first "page", etc.
> 
> It is probably simplest to implement this behavior if you first
> reshape X using the equivalent of
> 
>   sz = size (X);
>   X = reshape (X, sz(1), prod (sz(2:end)));

This won't work since you have to iterate over the first non-singleton
dimension of the array. Thus the above code wouldn't work for the 
case "a=rand(1,2,2);for i=a; disp(i); end;" 

Basically the patch I submit in the previous mail implements 

sz = size(X);
dim = 1;
for i=1:ndims(X)
 if (sz(i) > 1)
   dim = i;
   break;
  endif
endfor
X = reshape (X, sz(dim), prod(sz((dim+1):end)));

Before iterating over the array. In fact it uses an idx_vector and the
ArrayN<T>::index function to extract parts of the array in the loop. It
works with all array objects including row and column vectors.

> 
> I think this will work in all cases, not just for N-d arrays.  You can
> add a special case for row vectors for speed.  I think liboctave
> already offers a way to extract columns from Matrix objects.  Perhaps
> that functionality should be moved to the Array base class so that it
> can also work for Cell objects.


Effectively the code needed is already in Array<T>::index..

Cheers
David

-- 
David Bateman                                address@hidden
Motorola CRM                                 +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 1 69 35 77 01 (Fax) 
91193 Gif-Sur-Yvette FRANCE

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary