Re: Class of User Types [Was: Moving code from octave-forge to octave [Was: polyderiv problem?]]

[David Bateman]

Paul Kienzle wrote:

> On Feb 25, 2005, at 4:20 AM, David Bateman wrote:
>
> > > How do you save and load such beasts from files?
> > >
> > > Does typeinfo give more complete information?
> > >
> > > If the types really are not compatible, then the string names should 
> > > not match.
> > >
> > > Maybe class should return "GF(2)" if it is GF(2), and octave can 
> > > pass the
> > > extra info in the parentheses to the zeros constructor.
> > >
> > > - Paul
> > Paul,
> >
> > load/save is no problems, I just save which extension field of GF(2) 
> > I'm in an the primitive polynominal, together with the class. Its 
> > just another piece of information associated with the variable.
> >
> > As for having class return all of the information on the type, I 
> > really hope your kidding. The problem is then that there are a huge 
> > number of different primitive polynomials available as we take larger 
> > extension fields of GF(2). For example GF(2^15) has 1800 different 
> > possible primitive polynomials. The problem is even worse for fixed 
> > point types, where each element of a matrix can have a different 
> > representation and each of those between 0 and 32 bits in the 
> > representation of the integer and decimal parts of the elements.
> >
> > The implication that class returns the full information to define the 
> > type, is that there is a completely different type for all possible 
> > galois fields and fixed point matrices. Even if there were seperate 
> > types for each of the possibilities, imagine the number  of  
> > different versions of the operators I'd need to implement, 
> > particularly for the fixed point types. In all practical senses this 
> > is not feasible.
> >
> > For the fixed point types, having different representations doesn't 
> > make them incompatiable. All aritmetic operators can still be 
> > performed. However the same can not be said for galois fields. So I 
> > think for galois fields we just have to live with the return string 
> > of class being the same for incompatiable galois fields...
>
>
>
> I was assuming a string containing 'class(subclass)' where octave used
> 'class' to determine which type to invoke and the type used 'subclass'
> to decide what to do.  But no I'm not taking this idea very seriously.
>
> Another alternative is to delay recording the primitive polynomial type
> until something is actually assigned to the matrix, and from then on 
> check
> that it is consistent.  Would that work for you?
Err, I have to admit that you can do

gf(floor(8*rand(4,4)),8) + ones(4,4)

As the galois/matrix operations are already taken into account with the 
matrix promoted the a galois field. However

a = zeros(4,4); a(:,1) = gf(floor(8*rand(4,1)),8)

won't promote a, but leave it as a matrix, and so the implementation of 
triu won't work. I suppose an alternative for the galois fields is to 
modify the assign function to promote the matrix to a galois field as 
well... This might also work for the fixed point operators.

> Or maybe something more generic, such as array(exemplar,dims) so that
> dispatch will work.  Also, this makes more sense for some types such
> as 'cell' for which I was assuming the zeros function would create a
> set of empty cells.
>
> Then triu would be:
>
>    z = array(X,size(X));
>    for i=1:columns(X), z(1:i,:) = X(1:i,:); end
>
>
> Otherwise, zeros(dims,'auto',exemplar) does do what we want assuming
> the class registers a zeros constructor.
I don't see that we need an explicit zero constructor,

octave_value retval = args(0);
retval = retval.resize(dim_vector(0,0)).resize(args(0).dims());

does exactly the right.

Regards
David

--
David Bateman                                address@hidden
Motorola Labs - Paris                        +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