
Ublas : 
Subject: Re: [ublas] todo list
From: Matwey V. Kornilov (matwey.kornilov_at_[hidden])
Date: 20100724 17:08:42
Not the full list but the problems that were already mentioned.
1.a. Interfaces. IMO the first approach is to implement a set of most
commonused forms and the second one is to implement general and complex
form that can be tuned in different ways by means of metatemplate magic.
1.b. Types handling. For instance: a singular matrix is diagonal, should we
use banded_matrix<T> as return type?
2. Preconditions handling and error handling. Some of preconditions are pure
mathematical (for instance: positive definiteness or orthogonal property)
and we can't check them directly. We need a way to handle errors which are
raised during running an algorithm.
3. Rounding errors handling. For instance: matrices with bad pseudorank.
Imagine that your program have to make decision based on the rank of matrix
but it can't due to rounding errors.
I think the initial positive definiteness can be violated by rounding errors
in Cholesky decomposition. Then we get an runtime error where it shouldn't
be. (?)
Nasos Iliopoulos wrote:
>
> Matwey is very true on both of those and we need to have some discussions.
> Do you think you can compile a list of items that need to be addressed so
> that we can be ahead of the problems to come?