help-gsl
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Help-gsl] new NLS interface


From: Patrick Alken
Subject: Re: [Help-gsl] new NLS interface
Date: Thu, 8 Sep 2016 12:07:09 -0600
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.2.0

Hello,

  Underneath the trust region solver, there are several methods to
choose from, specified by the "trust region subproblem" variable "trs"
in the gsl_multifit_nlinear_parameters. There is LM, dogleg,
double-dogleg, and 2D subspace. You need to set the variable:

fdf_params.trs =
gsl_multifit_nlinear_trs_{lm,lmaccel,dogleg,ddogleg,subspace2D}

to select a different method. There are several example programs
illustrating the different methods.

The top-level method gsl_multifit_nlinear_trust is there to allow
possible expansion to other methods, such as line-search.

Currently, the new interface does not directly support Tikhonov
regularization, but I do plan to add this soon. In the mean time you can
use the old interface, or implement it directly in your residual and
Jacobian functions.

Patrick

On 09/08/2016 06:07 AM, viktor drobot wrote:
> Hello everyone! Am I right that in new NLS implementation there is only one
> solver - gsl_multifit_nlinear_trust? But in documentation there is example
> where the solver inits with gsl_multifit_nlinear_lm type of solver... I
> can't find more gsl_multifit_nlinear_type predefined variables...
> And one more question: where is the Tikhonov's regularization procedure
> now? Is it deprecated?
>




reply via email to

[Prev in Thread] Current Thread [Next in Thread]