Re: NonlinearRegress
- To: mathgroup at smc.vnet.net
- Subject: [mg49966] Re: [mg49915] NonlinearRegress
- From: "Janos D. Pinter" <jdpinter at hfx.eastlink.ca>
- Date: Sun, 8 Aug 2004 05:37:52 -0400 (EDT)
- References: <200408060709.DAA21401@smc.vnet.net>
- Sender: owner-wri-mathgroup at wolfram.com
Chris,
in general, you will need to use a global search/optimization procedure to
guarantee the satisfaction of parameter range constraints, as well as to
find the best numerical fit (expressed in a given norm). Consult e.g. item
140 in my list of publications at www.pinterconsulting.com: in that article
my MathOptimizer package is used to solve model calibration problems
globally, as opposed to built-in local model fitting functionality. You
could also try to use the built-in function NMinimize to attain the same
objective.
Frank Kampas and I have done some systematic numerical comparisons among
global solvers for Mathematica - in solving some rather difficult global
optimization problems - which we plan to make available in published form.
Regards,
Janos Pinter
At 04:09 AM 8/6/2004, you wrote:
>I've been using the NonlinearRegress function to extract parameter values
>from experimental data, with mixed success. While normally the extraction
>works well, it seem to fail spectacularly when it does - extracting values
>several orders of magnitude outside the specified range.
>
>Is there any way to guarantee that the extracted parameters fall within the
>range given for them, even if those parameters do not represent the best
>global fit ?
>
>I have been using the following options -
>
>NonlinearRegress[<data>, <expected function>, <variables>, {{<param1>,<
>startvalue>,< min>, <max>}. {<param2>, ....}}, MaxIterations->100000,
>Method->Automatic, RegressionReport-> BestFitParameters]
>
>to extract 3-4 parameters from ~350 data points.
>
>Thanks for your help,
>
>Christopher O'Brien
>University of Queensland
- References:
- NonlinearRegress
- From: "Chris O'Brien" <obrien@itee.uq.edu.au>
- NonlinearRegress