[Insight-users] ITK Dashboard: Our plan for fixing the problem with NonUniformBSpline

[Bradley Lowekamp]

Hello Luis,

I agree with your solution here. It seems much better to use a specific methods to solve the equation where the erroneous or unstable conditions can be better explicitly handled. 

However, I am still left with the question what is causing the the compiler error. Is this a real problem in VNL? I suppose we really are not specifically testing vnl, so we can program around these issues.
 
Brad

On Apr 29, 2010, at 6:08 PM, Luis Ibanez wrote:

> 
> The compilation problem that the (previously untested)
> NonUniformBSpline class is displaying in many platforms:
> 
> http://public.kitware.com/dashboard.php?name=itk
> 
> 
> arises from the expression in line 313:
> 
> The following code:
> 
>   vnl_matrix<double> B = 
>     vnl_matrix_inverse<double>( N_matrix.transpose() * N_matrix) * N_matrix.transpose() * data_matrix;
> 
> 
> is trying to solve the linear system
> 
>              X  = N * B
> 
> by doing:
> 
>         B = ( N' * N )^-1  *  N'  * X
> 
> This could be better done by using of 
> of the linear solvers in VXL. For example
> the LSQR one that was fixed recently.
> 
> 
> So, the suggested plan to solve this
> compilation problem is to replace that
> line 313 with a call to the vnl linear 
> system class and then to the vnl_lsqr
> solver.
> 
> This should be happening later today...
> 
> 
>     Luis
> 
> 
> <ATT00001..txt>

========================================================
Bradley Lowekamp  
Lockheed Martin Contractor for
Office of High Performance Computing and Communications
National Library of Medicine 
blowekamp at mail.nih.gov


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20100430/146d8087/attachment.htm>