optimset {neldermead} | R Documentation |
Configures and returns an optimization data structure.
Description
This function creates or updates a list which can be used to modify
the behaviour of optimization methods. The goal of this function is to manage
the options
list with a set of fields (for example, 'MaxFunEvals',
'MaxIter', etc...). The user can create a new list with empty fields or
create a new structure with default fields which correspond to a particular
algorithm. The user can also configure each field and set it to a particular
value. Finally, the user passes the list to an optimization function so
that the algorithm uses the options configured by the user.
Usage
optimset(method = NULL,...)
Arguments
method |
If provided, the |
... |
Additional arguments which would be included in the options output
if the |
Details
Most optimization algorithms require many algorithmic parameters such as the
number of iterations or the number of function evaluations. If these
parameters are given to the optimization function as input parameters, this
forces both the user and the developer to manage many input parameters. The
goal of the optimset
function is to simplify the management of input
arguments, by gathering all the parameters into a single list.
While the current implementation of the optimset
function only supports
the fminsearch
and fminbnd
function, it is designed to be
extended to as many optimization function as required. Because all
optimization algorithms do not require the same parameters, the data structure
aims at remaining flexible. But, most of the time, most parameters are the
same from algorithm to algorithm, for example, the tolerance parameters which
drive the termination criteria are often the same, even if the termination
criteria itself is not the same.
Optimization parameters that are returned by the optimset
function and
that can be defined in ...
are the following:
- Display
The verbose level. The default value is 'notify'. The following is a list of available verbose levels.
- 'off'
The algorithm displays no message at all.
- 'notify'
The algorithm displays message if the termination criteria is not reached at the end of the optimization. This may happen if the maximum number or iterations of the maximum number of function evaluations is reached and warns the user of a convergence problem.
- 'final'
The algorithm displays a message at the end of the optimization, showing the number of iterations, the number of function evaluations and the status of the optimization. This option includes the messages generated by the 'notify' option i.e. warns in case of a convergence problem.
- 'iter'
The algorithm displays a one-line message at each iteration. This option includes the messages generated by the 'notify' option i.e. warns in case of a convergence problem. It also includes the message generated by the 'final' option.
- FunValCheck
A logical flag to enable the checking of function values.
- MaxFunEvals
The maximum number of evaluations of the cost function.
- MaxIter
The maximum number of iterations.
- OutputFcn
A function which is called at each iteration to print out intermediate state of the optimization algorithm (for example into a log file).
- PlotFcns
A function which is called at each iteration to plot the intermediate state of the optimization algorithm (for example into a 2D graphic).
- TolFun
The absolute tolerance on function value.
- TolX
The absolute tolerance on the variable x.
- nbMatch
Specific to Box method: the number of consecutive times the
TolFun
criteria must be met to terminate the optimization.- boundsAlpha
Specific to Box method: the parameter used to project the vertices into the bounds in Box's algorithm
- boxScaling
Specific to Box method: the scaling coefficient used to scale the trial point for function improvement or into the constraints of Box's algorithm
- alphaMin
Specific to Box method: the minimum value of alpha when scaling the vertices of the simplex into nonlinear constraints in Box's algorithm
Output and plot functions The 'OutputFcn' and 'PlotFcns' options accept as argument a function (or a list of functions). In the client optimization algorithm, this output or plot function is called back once per iteration. It can be used by the user to display a message in the console, write into a file, etc... The output or plot function is expected to have the following definition:
myfun <- function(x, optimValues, state)
where the input parameters are:
- x
The current point estimate.
- optimValues
A list which contains the following fields:
- funccount
The number of function evaluations.
- fval
The best function value.
- iteration
The current iteration number.
- procedure
The type of step performed. This string depends on the specific algorithm (see
fminsearch
for details).
- state
the state of the algorithm. The following states are available:
- 'init'
when the algorithm is initializing,
- 'iter'
when the algorithm is performing iterations,
- 'done'
when the algorithm is terminated.
Value
Return a list with the following fields: Display, FunValCheck, MaxFunEvals, MaxIter, OutputFcn, PlotFcns, TolFun, TolX, nbMatch, boundsAlpha, boxScaling, and alphaMin.
Author(s)
Author of Scilab neldermead module: Michael Baudin (INRIA - Digiteo)
Author of R adaptation: Sebastien Bihorel (sb.pmlab@gmail.com)
See Also
optimset.method
,fminsearch
, fminbnd
Examples
optimset()
optimset(Display='iter')
optimset(method='fminbnd')