| Re: Powell's ghost [message #7187] |
Fri, 04 October 1996 00:00 |
Mark Fardal
Messages: 51
Registered: October 1995
|
Member |
|
|
As the function I need to minimize gets slower and slower, I've been
looking more closely at what POWELL does. What a piece of garbage.
The ITMAX and FTOL keywords are both ignored completely, and so it
merrily goes off and minimizes my function to 1 part in 10^12 when
I told it 10^-4 error was all right. Peter Mason points out to me
that this is "core IDL" (thanks Peter), so I can't even get to the
code to fix it.
The evaluated function is the main source of my current fitting
problems, but the excess accuracy in POWELL is sure slowing me down.
Does anyone have a properly implemented version of this or another
multidimensional minimization routine? I can only use ones without
analytic derivatives.
Thanks,
Mark Fardal
University of Colorado
|
|
|
|
| Re: Powell's ghost [message #7192 is a reply to message #7187] |
Fri, 04 October 1996 00:00  |
hto
Messages: 15
Registered: April 1996
|
Junior Member |
|
|
Mark Fardal <fardal@shapley.colorado.edu> wrote:
> Hi,
> I've been using the Numerical Recipes routine POWELL to do some parameter
> fitting. At first it worked nicely, but lately I've been having...
> Mark Fardal
> University of Colorado
Visit http://cfata2.harvard.edu/nr/
I believe that you can download the entire Numerical Recipes book or
any sub-section thereof (not the code). You can probably find the
section on the Powell method and also a description of other methods.
The book has the code embedded in the text. (Although I haven't
personally checked out whether that part of the text is available for
download) If you are going to be using Numerical Recipes a lot I
would HIGHLY recommend the text. It is very nicely written -- well
worth the price. I have purchased the original and the updated C
versions and I will purchase every new version that comes out. I also
purchased the C code. It has been worth every dime.
Howard Onishi
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
