Re: New sparse class

[David Bateman]

Daprès Paul Thomas <address@hidden> (le 02/11/2004):
> 
> >Also the "./" operator
> > for two sparse matrices doesn't include the NaN values as implied by the
> > division by the zero elements of the sparse matrix. This is how Matlab
> > treats this even though it is inconsistent with full matrices.
> > 
> 
> R12 seems to do the right thing
> 
> >> b = sparse( [1 2] , [2 1] , [5 0] , 2 , 2)
> b = 
>   (1,2)        5
> >> a = sparse( [1 2] , [2 1] , [5 5] , 2 , 2)
> a =
> 
>    (2,1)        5
>    (1,2)        5
> >>  a ./ b
> Warning: Divide by zero.
> ans =
>    (1,1)      NaN
>    (2,1)      Inf
>    (1,2)        1
>    (2,2)      NaN
> >> full(a) ./ full(b)
> Warning: Divide by zero.
> ans =
>    NaN     1
>    Inf   NaN

Err, ok, then its a bug on my part... In fact the operation I meant was
".^", which Matlab R12 when used with a matrix and a scalar gives.

>> a = speye(2) .^ 0       

a =

   (1,1)        1
   (2,2)        1

>> eye(2) .^ 0

ans =

     1     1
     1     1


There are quite a few strange and undocumented misfeatures of this kind
in the Matlab sparse class.

Regards
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