Re: Carlo : GsoC Numerical Project

[c.]

On 2 May 2013, at 11:35, address@hidden wrote:

> Message: 7
> Date: Thu, 2 May 2013 10:19:55 +0200
> From: othman darraz <address@hidden>
> To: address@hidden
> Subject: Carlo : GsoC Numerical Project
> Message-ID:
>       <address@hidden>
> Content-Type: text/plain; charset="iso-8859-1"
> 
> Hello Carlo,
> 
> As I am applying for one of Octave' Numerical Projects. On the IRC, JordiGH
> recommend that I should ask you about it. Can you please give me some
> guidelines or advises to improve my proposal. Did I understand well what it
> is supposed to do in this project?
> 
> I recently sent a email fleshing out more detail what i want to do in this
> project:
> 
> My objective would be to code incomplete LU and Cholesky factorization for
> Sparse Matrix in CCS format using m-Scripting
> 
> I want to begin with incomplete LU with no fill ( 1 week )then with all
> degrees of fill ( which seems in theory harder 2 weeks) then i will work on
> Cholesky factorization ( 1 week ) and finally interfacing Octave to ITSOL (
> I don't really have an idea how it 's could be done right now )
> I didn't write any code for the moment but i have some m-script i found on
> this discussion.
> 
> http://octave.1599824.n4.nabble.com/Ilu-function-td4648677.html
> 
> Particularly, full2ccs.m should be helpful
> 
> 
> Thank you for your time


Hi Othman,

Sorry I hadn't noticed your proposal on melange I now added myself as a 
possible mentor for it.

As for your questions:

1) I don't think working with m-files is a viable solution in this case, unless 
it is intended 
for preparing a prototype implementation to be later reimplemented in C++, but 
if we plan to have
the Octave code implemented as a wrapper around an existing library it makes 
even less sense to have an 
m-file prototype.

2) bear in mind that in this project full matrices should be avoided as much as 
possible as iterative 
methods and thus preconditioners are only useful for very large and sparse 
matrices. 
Getting CCS vectors format from an Octave SparseMatrix in an m-file is 2-liner:

[i, j, v] = find (A);
[~, j] = unique (j, "first");

to make your application stronger I'd recommend you look into building Octave 
from sources and start looking 
at the SparseMatrix class so you can add more technical details to the C++ part 
of your workplan.
c.