Prev: matrix operations
Next: Get time during ode113 solver
From: Peter Knief on 7 Jun 2010 07:47
Hi Guys,

I'm using Matlab 2008b with the curve fitting toolbox, and only some data (all have the same structure) the 'Levenberg-Marquardt' - algorithm throws :

??? Undefined function or variable "pfull".

Error in ==> fit at 437
errstr = handleerr( errid, errmsg, suppresserr );


I couldn't find anything about is, who can help me / What can possibly be the cause?

thanks in advance
From: Jared on 28 Jun 2010 15:00
"Peter Knief" <peter(a)knief.de> wrote in message <huim7n$6ob$1(a)fred.mathworks.com>...
> Hi Guys,
>
> I'm using Matlab 2008b with the curve fitting toolbox, and only some data (all have the same structure) the 'Levenberg-Marquardt' - algorithm throws :
>
> ??? Undefined function or variable "pfull".
>
> Error in ==> fit at 437
> errstr = handleerr( errid, errmsg, suppresserr );
>
>
> I couldn't find anything about is, who can help me / What can possibly be the cause?
>
> thanks in advance


I am having the same issue using the trust region algorithm and only on certain sets of data. Is there a solution for this?
From: Jared on 30 Jun 2010 14:32
"Peter Knief" <peter(a)knief.de> wrote in message <huim7n$6ob$1(a)fred.mathworks.com>...
> Hi Guys,
>
> I'm using Matlab 2008b with the curve fitting toolbox, and only some data (all have the same structure) the 'Levenberg-Marquardt' - algorithm throws :
>
> ??? Undefined function or variable "pfull".
>
> Error in ==> fit at 437
> errstr = handleerr( errid, errmsg, suppresserr );
>
>
> I couldn't find anything about is, who can help me / What can possibly be the cause?
>
> thanks in advance

After speaking with Tech support, I have learned that this issue is an edge-case bug. This can happen when the fitting requires something like 605 function evaluations when the maximum is set to 600. Function evaluations come in batches so sometimes it requires more than you allow it. If this happens then the variable "pfull" is never defined which spits out an error, in my case in the file cfrobnlinfit.m.

However, this basically means that if you are running a program where a number of data sets are being modelled (in my case over 1000), it is very hard to set the right number of maximum function evaluations. There could always potentially be a data set that requires just a few more. So I am hoping Mathworks will provide a cleaner fix to the issue.

Hope that helps.
 | 
Pages: 1
Prev: matrix operations
Next: Get time during ode113 solver