Re: Optimisation functions missing? (fwd)

[heberf]

Your friends comments were interesting and (I think) correct.  It all comes 
down 
to the following: there is no easy answer the question "What is the best 
optimization method".  For many computational computer types this is 
frustrating 
but it's true.  Optimization is the black art of numerical computing.  In 
practice what you do is try something and if it doesn't work you tweek the 
starting values and change some parameters and if that doesn't work you use a 
different method.

What this means for Octave is that we are (probably) never going to have a 
function called nlp that does general nonlinear programming.  We should code up 
several different functions and put them in a toolbox.  

On a slightly more positive note let's be practical here.  Just because we 
can't 
prove that a certain method is best doesn't mean we shouldn't use it.  Best 
just 
means fastest.  If you can wait a few minutes more then we can use one of the 
many existing methods and for some problem is will be best but not for others.  
Either way we will still get the solution.

Free software really frees us from a lot of these questions.  If we were 
writing 
a commercial package we would stress about this because we would be worried 
about our package becoming obsolete.  With GPL's software if someone figures 
out 
a better way then they can change the code and submit it.  That way we all win. 
 
The point is let's gets started.  The first improvement is easy because 
something is better than nothing.  Let's get some first drafts of code up and 
tinker.

Heber