Thesis DISEÑO DE UNA ESTRATEGIA PARA OBTENER CONOCIMIENTO EN UN SINTONIZADOR DE PARÁMETROS
Date
2022-11
Authors
TORO VEGA, DANIEL ALEJANDRO
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
La toma de decisiones es un aspecto cotidiano tanto para las personas como para las grandes corporaciones. Una gran cantidad de decisiones pueden ser modeladas como un problema de optimización. Los problemas de optimización son problemas en donde se debe decidir cómo utilizar los recursos disponibles, de tal manera de lograr un objetivo. Para resolver problemas de optimización es frecuente el uso de metaheurísticas, algoritmos que son capaces de obtener soluciones de buena calidad en un tiempo razonable. Sin embargo, el uso de metaheurísticas suele introducir el problema de determinar los valores adecuados para sus parámetros. Este problema se le conoce como el problema de seteo de parámetros (PSP). Para abordar este problema se han propuesto métodos de sintonización y control de parámetros. En esta memoria se trabajó con el sintonizador de parámetros Evolutionary Calibrator (EVOCA) y se propuso una nueva versión. Esta nueva versión utiliza un modelo de clasificación, Naive Bayes Classifier (NBC) para determinar si el operador de mutación se debe utilizar en otro parámetro. A esta nueva versión se le llamó EVOCA+NBC. Se obtuvo evidencia experimental donde EVOCA+NBC obtiene mejores resultados para las configuraciones en términos de calidad y tiempo de ejecución. Además, se vio que es posible obtener configuraciones equiparables a EVOCA en menos evaluaciones.
Decision making is a common aspect for people and even big corporations. A wide range of decisions can be modeled as an optimization problem. Optimization problems are problems where one must decide how to use available resources to achieve a goal. To solve this kind of problems it is common to use metaheuristics, algorithms that can produce good quality solutions in reasonable time. However, the use of metaheuristics introduces the problem to determine the adequate values of its parameters. This problem is known as the Parameter setting problem (PSP). To tackle this problem parameter control and tuning methods have been proposed. In this work the parameter tuner Evolutionary Calibrator was used and a new version was proposed. This new version utilizes a classification model, Naive Bayes Classifier (NBC), to determine if the mutation operator should be used on another parameter. The new version was named EVOCA+NBC. Experimental evidence was found that shows that EVOCA+NBC can obtain better results, in terms of solution quality and execution time. Additionally, it was seen that it is possible to obtain comparable configurations to EVOCA in less evaluations.
Decision making is a common aspect for people and even big corporations. A wide range of decisions can be modeled as an optimization problem. Optimization problems are problems where one must decide how to use available resources to achieve a goal. To solve this kind of problems it is common to use metaheuristics, algorithms that can produce good quality solutions in reasonable time. However, the use of metaheuristics introduces the problem to determine the adequate values of its parameters. This problem is known as the Parameter setting problem (PSP). To tackle this problem parameter control and tuning methods have been proposed. In this work the parameter tuner Evolutionary Calibrator was used and a new version was proposed. This new version utilizes a classification model, Naive Bayes Classifier (NBC), to determine if the mutation operator should be used on another parameter. The new version was named EVOCA+NBC. Experimental evidence was found that shows that EVOCA+NBC can obtain better results, in terms of solution quality and execution time. Additionally, it was seen that it is possible to obtain comparable configurations to EVOCA in less evaluations.
Description
Keywords
ALGORITMOS COMPUTACIONALES , TOMA DE DECISIONES , INVESTIGACION OPERATIVA