[Insight-developers] Re: Kernel transforms. K symmetric?

[Luis Ibanez]

Jim,

That sounds great,

If only one additional method is needed the
changes will be minimal. Probably the overhead
of filling the diagonal is also negligible
compared to the cost of computing the rest
of the K matrix.


   Luis


-----------------------------------------

Miller, James V (Research) wrote:

> Luis, 
> 
> The way I have it now, there is a new method called
> SetStiffness(double) on a kernel transform (have not 
> determined whether this will go in the KernelTransform
> class or just those kernel transforms that can be formulated 
> like this).  Anyway the default stiffness is 0.0 and results
> in an interpolating spline.  So a subclass in not needed.
> 
> For weighted splines, I think I can just add a SetWeights() method
> to associate a weight or covariance matrix with each the source points.
> Another option is to attach the weights to the PointSet defining the 
> source points.
> 
> There could be a mode as to whether the weights are to be used 
> or not.
> 
> Haven't got it all mapped out yet...
> 
> Jim
> 
> 
> 
>