[scilab-Users] fminsearchbnd

[Michaël Baudin]

Dear Carlos,

You will find in attachement a demo of neldermead with a simple
bound-constrained problem. It minimizes x(1)^2+x(2)^2
in the bounds [1 2]x[1 2] and find the minimum at (1,1).
Notice that the simplex goes toward the minimum bound,
degenerates, then performs a 1 dimensionnal optimization
on the min x bound.
There is a lot of blabla in the script, but it remains simple.

Best regards,

Michaël

Michaël Baudin a écrit :
> Dear Carlos,
>
> I assume that you are talking about the Matlab toolbox by John D'Errico.
> There is no fminsearchbnd in the current release, but there is the 
> neldermead
> component, which includes a Box algorithm which manages both
> bounds and general nonlinear constraints. In your console :
>
> -->help neldermead
>
> Look in the ?> Scilab Demonstrations> Optimization and Simulation> 
> neldermead/Box A
> The Box A problem is an optimization problem with bounds and nonlinear 
> constraints.
>
> You should have no problem with bounds, provided that the number of 
> variables is
> low : up to 10-20 is a maximum.
>
> A typical alternative is to use the unconstrained fminsearch and to apply
> a penalty when x is not in the bounds. A simple scaling parameter might
> do the trick, but the scaling obviously depends on the problem, that 
> is, depends
> on the function value.
>
> Best regards,
>
> Michael
>
> PS
> Notice that John D'Errico's function is based on variable 
> transformations, while
> Box's algorithm modifies the updating scheme of the simplex.
>
> Sébastien Bihorel a écrit :
>> Hi Carlos,
>>
>> I don't believe that scilab includes this function. One way to 
>> implement boundaries on parameters with fminsearch is to apply them 
>> directly in the function to be minimized (and in your custom output 
>> function, if you have one).
>>
>> Sebastien.
>>
>> On Tue, Feb 16, 2010 at 8:27 PM, Carlos Eduardo Celia de Lima 
>> <carlosedlima at gmail.com <mailto:carlosedlima at gmail.com>> wrote:
>>
>>     Dear Sci-Lab experts,
>>
>>     Is there a equivalent function to fminsearchbnd in scilab?
>>     I was not able to find it in the documentation.
>>
>>     If there is not, is there any workaround to apply boundaries to
>>     the variables in an optimization problem?
>>
>>     Thank you,
>>
>>     Carlos
>>
>>
>
>


-- 
Michaël Baudin
Ingénieur de développement
michael.baudin at scilab.org
-------------------------
Consortium Scilab - Digiteo
Domaine de Voluceau - Rocquencourt
B.P. 105 - 78153 Le Chesnay Cedex
Tel. : 01 39 63 56 87 - Fax : 01 39 63 55 94


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: neldermeadBounds.sce
URL: <https://lists.scilab.org/pipermail/users/attachments/20100217/501c5551/attachment.ksh>