Core Concepts
The BaumEvA library provides efficient and reliable methods for solving a variety of optimization problems, including conditional and unconditional optimization, as well as binary tasks, through the implementation of evolutionary algorithms.
Abstract
The report presents the results of comprehensive testing of the BaumEvA evolutionary optimization Python library. The library implements various evolutionary algorithms, including binary genetic algorithms, categorical genetic algorithms, and combinatorial genetic algorithms, to solve optimization problems.
The testing was conducted on both conditional and unconditional optimization problems, as well as binary tasks, to evaluate the effectiveness and reliability of the proposed methods. For the conditional optimization problem (CO1), the library was able to find the minimum with the given accuracy (0.0001) for the lower-dimensional problem (D10), but faced challenges in higher dimensions (D30, D50, D100) due to the limitations on the number of function evaluations.
For the unconditional optimization problems, the library demonstrated good performance in solving unimodal and multimodal functions, achieving the desired accuracy of 10^-8 in most cases. However, some multimodal functions, such as Rastrigin's function, required more computational resources to reach the optimum.
The testing of binary tasks, including OneMax, LeadingOnes, and Trap functions, showed excellent optimization results across different dimensions (D50, D100, D200, D500, D1000). The library was able to find the optimal solutions efficiently, with the number of function evaluations required increasing as the problem dimension increased.
The report also discusses the limitations of the library, such as the execution time for high-dimensional multimodal problems, and provides recommendations for choosing algorithm parameters and using the library to achieve the best results. Overall, the BaumEvA library is a powerful tool for solving a variety of optimization problems and can be effectively used in various fields of science and technology.
Stats
The minimum number of function evaluations (FEs) required to achieve the optimum for the binary tasks are:
OneMax function: 190 FEs (D50), 550 FEs (D100), 1378 FEs (D200), 4114 FEs (D500), 10927 FEs (D1000)
LeadingOnes function: 1207 FEs (D50), 5464 FEs (D100), 22348 FEs (D200), 177787 FEs (D500), 962983 FEs (D1000)
Trap function: 190 FEs (D50), 460 FEs (D100), 1144 FEs (D200), 4042 FEs (D500)
Quotes
"The library provides an efficient implementation of the genetic algorithm, in particular the binary genetic algorithm and with Gray codes."
"Testing has shown the high efficiency of the algorithms in solving problems of unconditional optimization of Boolean functions, unimodal continuous functions and conditional optimization."
"The library is highly scalable and can be used to solve problems of varying complexity and data volume, but sometimes with large dimensions and very small steps (10e-8) calculations can take a long time."