SCILAB PROGRAMS
FOR OPTIMIZATION
Laurent Dumas
All these programs are tested on the same example, the Rastrigin function with n parameters on an hypercube.
With minor changes, they can be used for any real function.
1. Deterministic methods:
A steepest
descent method with a backtracking linesearch: steepest.sci
A
BFGS method with a backtracking linesearch: BFGS.sci
The
Nelder Mead method: NelderMead.sci
The
Multi Direction Search algorithm: MDS.sci
2. Stochastic methods:
A
basic simulated annealing: SA.sci
A
basic real valued genetic algorithm: GA.sci
A
basic evolution strategy: ES.sci
An
ES with cumulative search adaptation: ES-CSA.sci , CSAES.m (idem
en Matlab)
A basic PSO algorithm: PSO.sci
A non-dominated sorted GA (without sharing): NSGA.sci
3. Hybrid method (stochastic+deterministic):
An
adaptive Hybrid method with an ES core : AHM-ES.sci
An
adaptive Hybrid method with a GA core : AHM-GA.sci
4. Surrogate model (stochastic+approximated evaluations):
A
surrogate model with a GA core: AGA.sci
5. Reference: my HDR dissertation