Re: NMinimize method in NonlinearModelFit not accepted!
- To: mathgroup at smc.vnet.net
- Subject: [mg108558] Re: NMinimize method in NonlinearModelFit not accepted!
- From: Darren Glosemeyer <darreng at wolfram.com>
- Date: Tue, 23 Mar 2010 04:23:39 -0500 (EST)
dg wrote: > I cannot set Method->"NMinimize" inside NonlinearModelFit in Mathematica 7 > Mac. I get the following error message: > > NonlinearModelFit::bdmtd : > Value of option Method -> NMinimize is not Automatic, > "Gradient",...,"NMinimize", or "LevenbergMarquardt" > > which clearly does not make sense. > > Everything works fine with Method->"LevenbergMarquardt". But then, Mathematica > is quite lousy at minimizing Chi squared... > The failure to accept "NMinimize" as a Method is a bug that has been fixed in the version currently under development. When using "LevenbergMarquardt", have you specified starting values for the parameters? Good starting values can be very important in local nonlinear optimization. Alternatively, constructing the sum of squares and minimizing directly with NMinimize could be done to get parameter estimates. You could potentially feed estimates obtained from NMinimize back into NonlinearModelFit as starting values to get diagnostics. Darren Glosemeyer Wolfram Research