Re: Choosing parameters.

[Jody Goldberg <jody gnome org>]

On Fri, May 24, 2002 at 11:29:02AM -0400, Daniel Carrera wrote:
> I suppose that it makes sense to try many points and see which one arrives
> at the lowest Chi Squared.
>
> Won't that be expensive though?  Also, how do we know how many points to
> try?  That number depends on the user-defined function and the size of the
> intervals chosen.
This is a standard limitation of solvers.  The user must do
additional testing of whatever result is produced.  There are no
hard and fast rules around it.
 
> Furthermore, wouldn't it still be good to help the user pick good ranges
> by seeing a plot?  Is this difficult?
Yes and no.
ploting can help visualize things with low dimensionality, but it
still requires user input to tune the parameters
    - what bounds ?
    - what step size ?

In the 1.1 development tree (when guppi is ready) we'll have direct
access to guppi.  Which should make it easier to throw up a plot if
desired.

Methodologies like simulated annealing attempt to get around this by
introduing a degree of randomness, but that introduces repeatability
issues.