The Math Forum

Search All of the Math Forum:

Views expressed in these public forums are not endorsed by NCTM or The Math Forum.

Math Forum » Discussions » Software » comp.soft-sys.matlab

Notice: We are no longer accepting new posts, but the forums will continue to be readable.

Topic: parameter normalization for optimization functions such as fmincon and fminsearch
Replies: 5   Last Post: Jun 26, 2009 8:35 AM

Advanced Search

Back to Topic List Back to Topic List Jump to Tree View Jump to Tree View   Messages: [ Previous | Next ]
Hanan Shteingart

Posts: 4
Registered: 6/25/09
Re: parameter normalization for optimization functions such as fmincon and fminsearch
Posted: Jun 25, 2009 1:23 PM
  Click to see the message monospaced in plain text Plain Text   Click to reply to this topic Reply

Thanks John for your quick reply.

Any idea of how to do what you've proposed (same order of magnitude for different parameters) ?
I am using a non-analytic function that measure the likelihood of a model to an observed measurements, so how can I tell in advance the derivation of the likelihood by each parameters? I don't have an explicit expression for the likelihood since the model is not stationary and is data dependent.


"John D'Errico" <> wrote in message <h20959$i6u$>...
> "Hanan Shteingart" <> wrote in message <h208cv$s5u$>...
> > Hi,
> > From other discussions I've seen here, it seems the TolX parameter in the option parameter to fmincon/fminsearch is absolute, which means those functions assume the same order of magnitude to all parameters. Is there any smart way to solve this? Should I normalized all parameters to be around [0 1] and re-normalize it within the subject function (which I want to minimize)?
> >

> Better than normalizing all parameters to be unity, would
> be to scale them to have derivatives that are all roughly
> the same order of magnitude.
> Lacking that, scaling them to be roughly the same
> magnitude may be reasonable.
> John

Point your RSS reader here for a feed of the latest messages in this topic.

[Privacy Policy] [Terms of Use]

© The Math Forum at NCTM 1994-2018. All Rights Reserved.