Comparative Evaluation of the Stochastic Simplex Bisection Algorithm and the SciPy.Optimize Module

• Autorzy: Christer Samuelsson
• Języki publikacji: EN

Abstrakty

EN

The stochastic simplex bisection (SSB) algorithm is evaluated against the collection of optimizers in the Python SciPy.Optimize module on a prominent test set. The SSB algorithm greatly outperforms all SciPy optimizers, save one, in exactly half the cases. It does slightly worse on quadratic functions, but excels at trigonometric ones, highlighting its multimodal prowess. Unlike the SciPy optimizers, it sustains a high success rate. The SciPy optimizers would benefit from a more informed metaheuristic strategy and the SSB algorithm would profit from quicker local convergence and better multidimensional capabilities. Conversely, the local convergence of the SciPy optimizers is impressive and the multimodal capabilities of the SSB algorithm in separable dimensions are uncanny.(original abstract)

Słowa kluczowe

EN

Algorithms

PL

Algorytmy

• Czasopismo: Annals of Computer Science and Information Systems
• Rocznik: 2015
• Tom: 5
• Strony: 573--578

Twórcy

autor

Christer Samuelsson

• German Research Center for Artificial Intelligence

Bibliografia

• M. Wahde, Biologically Inspired Optimization Algorithms. WIT Press, 2008.
• D. E. Goldberg and J. H. Holland, "Genetic algorithms and machine learning," Machine Learning, vol. 3, no. 2-3, pp. 95-99, 1988. doi: 10.1023/A:1022602019183
• J. Kennedy and R. Eberhart, "Particle swarm optimization," 1995. doi: 10.1109/ICNN.1995.488968
• M. Dorigo, V. Maniezzo, and A. Colorni, "The ant system: Optimization by a colony of cooperating agents," IEEE Trans. on Systems, Man, and Cybernetics, vol. 26, no. 1, pp. 29-41, 1996. doi: 10.1109/3477.484436
• C. Samuelsson, "The stochastic simplex bisection algorithm," in Procs. 15th Int.'l Conf. Computational Science, ser. ICCS/15. Elsevier, 2015. doi: 10.1016/j.procs.2015.05.215 pp. 855-864. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S1877050915010236
• scipy, "scipy.optimize," http://http://docs.scipy.org/doc/scipy/reference/ optimize.html

Identyfikatory

DOI

10.15439/2015F47