|
|
Re: Non-linear optimization
Posted:
Mar 4, 2013 10:36 PM
|
|
"Steven_Lord" <slord@mathworks.com> wrote in message <kh2ni9$9ar$1@newscl01ah.mathworks.com>...
>
>
> "Toan Cao" <toancv3010@gmail.com> wrote in message
> news:kh2m44$4eh$1@newscl01ah.mathworks.com...
> > Hi everyone,
> >
> > I have a question relating to non-linear optimization and hope to receive
> > your help!
> > If i have a cost function F(x) in general form. I mean it can not be
> > described in the form of F(x)= f(x)'.f(x) (where f(x)' is transposition of
> > f(x) ) as required by some methods such as Levenberg-Marquardt,
> > Gauss-Newton for finding a local minimum value.
> > If i want to use above two methods, what should i do ?
> > Do you know any method that i can use in this case?
>
> Do you have a MATLAB function that accepts a vector of parameters to be
> optimized and returns the value of your cost function? If so, you're okay.
> Most of the Optimization Toolbox and Global Optimization Toolbox solvers
> don't care what you do inside your objective function to obtain the values
> they seek (as long as your underlying function is smooth, for some of them)
> as long as you return the cost function values from your objective function.
>
> You could do calculations. You could retrieve information from disk. You
> could query an instrument or database. You could even prompt the user to
> INPUT the result of performing a physical experiment for a set of parameter
> values.
>
> --
> Steve Lord
> slord@mathworks.com
> To contact Technical Support use the Contact Us link on
> http://www.mathworks.com
Thanks for your reply, Steve Lord.
Can i ask you more about this problem ?
As i know, Levenberg-Marquardt, Gauss-Newton methods use f(x) to calculate the Jacobian matrix. If i just have F(x) ( not f(x) ), how does optimization solvers compute this matrix? Can you explain for me the way that the solvers implement ?
Actually, i need to understand more deeply and hope to modify somethings for my specific function.
Thanks in advance.
Toan
|
|
