OpenOpt - Optimization Software
OpenOpt is a web site for freely available optimization software. The following downloads are available
Vopt-2D a visual platform for testing optimizers
This visual graphics platform is an OpenSource project that benefits from parts of the OpenMaXwell code for computational electromagnetics. It is written in Fortran, using the QuickWin library, and compiled with the Intel Visual Fortran compiler. Vopt-2D supports graphics for visualizing the fitness landscape and the progress of various optimizers such as Genetic Algorithms (GA) and microGA, Evolution Strategy (ES), Particle Swarm Optimization (PSO), and Nelder-Mead (simplex). The user may define various fitness landscapes that provide different difficulties for the optimizer. Although only 2D graphics is available, higher dimensional parameter optimization may be studied. The ZIP file contains executables (for Windows), source code, and a short manual.
Evolution Strategies for real parameter optimization
Standard ES (SES) and a generation-free version for efficient parallel optimization (PES), packed with OpenMaXwell examples in a ZIP file.
Binary optimization with incomplete fitness tables for problems with short bit strings
Genetic Algorithms (GAs) are frequently used for binary optimizations. In engineering one often has rather short bit strings but very long computation times for evaluating the fitness values. Then, standard GAs are not suited and als microGAs are not efficient because they often recompute the fitness of a model that had been computed before. By maintaining a table that stores all fitness evaluations, one may avoid multiple fitness evaluations. Furhter speed-up may be obtained from a statistical analysis of the information contained in such incomplete fitness tables. The package contains several optimizers based on the microGA concept, on some sort of mutation-based binary Evolution Strategy (ES), etc. There is not much documentation, but the Fortran codes of the optimizers are rather simple. The ZIP file also contains several complete fitness tables (lookup tables) of pre-computed examples. These tables may be used for rapidly testing and comparing various optimizers and the influence of tuning parameters such as generation size, mutation rates, etc. The software is prepared to collaborate with the OpenMaXwell solver in a very similar way as SES for real parameter optimization. A parallel binary optimization strategy will be added later.
Responsible for this web page: Ch. Hafner, Computational Optics Group, IEF, ETH, 8092 Zurich, Switzerland
Last update 04.12.2014