feedback on OPTIM macro

[Carrico, Paul]

Dear All,

 

I 'm currently testing the optim macro in Scilab/FEA (external) solver FEA  and I would like here to share some feedbacks ...

 

First of all regarding the documentation, I noticed :

 

 

 

A) 'gc' and bounds use

'qn' -> BFGS algorithm

'gc' -> conjugate gradient

'nd' -> ?

>From the documentation  "Note that the conjugate gradient solver does not accept bounds on x. "

Whereas the example "Scilab function" uses bounds with 'gc' ... ???? 

// Upper and lower bounds on x and setting up the algorithm to 'gc'
[f,xopt,gopt]=optim(cost,'b',[-1;0;2],[0.5;1;4],x0,'gc')

 
B) bounds violated
I tried several syntaxes for the bounds (with or without the dot coma i.e vector or its transposed) and all the time I noticed negative values of x ...
 [fopt,optimized_parameters] = optim(optimization_BFGS,'b',[0;0],[2;2],initial_parameters','ar',nap,iter,epsg,imp=-1);
 Is there something wrong ?
 
 B) derivative use
 I noticed the strong influence of the h parameters (for the step) as well as the order parameter (order 2 by default) :
ü       If h is too small è a lot of evaluations
ü       If h is too high è loops quickly stop far from the optimized values 
 order is essentially used for the function evaluation ; I tried different values of "order" thinking in a better function evaluation (for a better gradient calculation) ... but the default value seems to be the best one.
 
  
C) Other algorithms
 I tested on a basic Rosenbrock function other order 0 method such as Genetic Algorithm or Simulated annealing but many function evaluations are necessary (more than a thousand) .... but probably the parameters have to be optimized themselves (Population size / number of mutations and so on ) ... 
 since one evaluation need one FE modelling (i.e. the longest calculation duration), I'm not sure they are the most efficient for my application ... finally the Nelder-Mead method is probably one of the best !
 
 
Regards
 
Paul

 


--------------------------------------------------------------------------------


Le présent mail et ses pièces jointes sont confidentiels et destinés à la personne ou aux personnes visée(s) ci-dessus. Si vous avez reçu cet e-mail par erreur, veuillez contacter immédiatement l'expéditeur et effacer le message de votre système. Toute divulgation, copie ou distribution de cet e-mail est strictement interdite.

This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error, please contact the sender and delete the email from your system. If you are not the named addressee you should not disseminate, distribute or copy this email.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/users/attachments/20110131/00b43604/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: diary_OPTIM_H_0p1_bounded.dat
Type: application/octet-stream
Size: 49416 bytes
Desc: diary_OPTIM_H_0p1_bounded.dat
URL: <https://lists.scilab.org/pipermail/users/attachments/20110131/00b43604/attachment.obj>