pwSetOptimizer


 oldValue = pwSetOptimizer(optimizer, showInfo)

 Sets the optimizer. Returns the number of the currently used optimizer
 (before setting to the new optimizer).

 

Arguments for pwSetOptimizer

 optimizer    1 or 'linesearch'
              2 or 'trustregion'
              3 or 'genetic'
              4 or 'annealing'
              5 or 'smarquardt'
              6 or 'ssmlsqnonlin'
              7 or 'ssmdn2fb'

 showInfo     If true, information about the changed optimizer
              is displayed in the command window. Default = true.

 

Requirements and properties

 

linesearch

 Multidimensional unconstrained nonlinear minimization (Nelder-Mead).
 Available within standard Matlab.
 Deterministic, slow, inaccurate.
 Does not support lower and upper limits for fitted parameters.

 

trustregion

 Subspace trust-region method based on interior-reflective Newton.
 Requires the optimization toolbox.
 Deterministic, fast, accurate.
 Supports lower and upper limits for fitted parameters.

 

genetic

 Genetic algorithm.
 Requires the genetic algorithm toolbox.
 Non-deterministic, very slow, usually not accurate with default settings.
 Requires experience with and tuning of optimization settings.
 Supports lower and upper limits for fitted parameters.

 

annealing

 Simulated annealing (ASAMIN).
 Included in PottersWheel (requires successful compilation of C MEX file,
 which is done during installation with pwInstall).
 Non-deterministic, slow, does not suffer from local minima, mediuam accuracy.
 Supports lower and upper limits for fitted parameters.
 Should be applied in combination with (i.e. before) trustregion,
 see pwFitBoost.

 

smarquardt

 Approximate Gauss-Newton with Levenberg-Marquardt damping and
 successive updating of Jacobian approximation, Hans Bruun Nielsen.
 Included in PottersWheel.
 Does not support lower and upper limits for fitted parameters.

 

ssmlsqnonlin

 Scattersearch combined with trustregion, Banga et al.
 Requires the optimization and SSM toolbox,
 http://www.iim.csic.es/~gingproc/ssmGO.html
 Combination of stochastic and deterministic optimization.
 Supports lower and upper limits for fitted parameters.

 

ssmdn2fb

 Scattersearch combined with N2FB, Banga et al.
 Requires the SSM toolbox, http://www.iim.csic.es/~gingproc/ssmGO.html
 Combination of stochastic and deterministic optimization.
 Supports lower and upper limits for fitted parameters.


See also

pwFit