Re: [Boost-Users] ublas - newbie question and comment

From: jhr.walter_at_[hidden]
Date: 2003-01-06 04:45:34

----- Original Message -----
From: Mike Hayford
To: Boost-Users_at_[hidden]
Sent: Tuesday, December 31, 2002 8:20 AM
Subject: [Boost-Users] ublas - newbie question and comment

> I discovered uBLAS a week ago and am very impressed by the design and
> implementation of the library.  To learn the library I have been porting a
few non-linear
> least squares codes from FORTRAN into uBLAS.  A real hinderance is the
> documentation.  There is no user's guide or set of example code to orient
the new user to > uBLAS.

OK.

> Some of the most valuable features are undocumented - such as
scalar_vector,
> vector_indirect, and matrix_indirect.  The widespread adoption of uBLAS by
the C++
> numeric community would be more rapid if these omissions were addressed.

OK.

> I would be willing to write up some of the *simple* examples I have done,
if people
> think it would be useful.  If my implementations can be improved, that
would be even
> better.  I'm new to Boost and haven't yet completely RTFM'd the
requirements for
> Boost doc contributions so point me in the right direction if I'm going
astray here.
>
> Having said that, I have a specific problem I'm trying to solve.  I'm
looking for the best
> "uBLAS-style" solution - I can do an element-by-element brute force
solution.
>
> Problem: write the product of a diagonal matrix and a vector when the
diagonal matrix is > stored as a vector.
>
> So what I want to do figuratively is:
>
> int n = 5;
>
> vector<double> x(n), y(n), diag(n);
>
> diagonal_matrix<double> dMat(n);
>
> dMat = diag;
>
> x = prod(dMat, y);
>
> Is there a matrix adaptor of some sort that can fill the role of
diagonal_matrix<> above?  > I looked in the header files and didn't see
anything obvious.

I'd use a banded_matrix with number of upper and lower diagonals set to
zero. Another strategy could be to add an expression tree node class for
this operation based on vector_binary.

Regards

Joerg