Experimentos Computacionais com Implementações de Conjunto por Endereçamento Direto e o Problema de Conjunto Independente Máximo

• Ano de defesa: 2013
• Autor(a) principal: Santos, Marcio Costa
• Orientador(a): Corrêa, Ricardo Cordeiro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Mapas de Bits; Conjunto independente máximo; Paralelismo de bits
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/13356

The use of bit vectors is a usual practice for represent sets by direct addressing with the aim of reduce memory consumed and improve efficiency of applications with the use of bit parallel techniques. In this text, we study implementations for represent sets by direct addressed. The basic structure in this implementations is the bit vector. Besides that basic implementation, we implement two variations also. The first one is a stratification of the bit vector, while the second uses a hash table. The operations linked to the implemented structure are include and remove an element and the union and intersection of two sets. Especial attention is given to the use of bit parallel in this condition. The implementation of the different structures in this work use an base interface and a base abstract class, where the operations are defined and the bit parallel is used. An experimental comparative between this structures is carry out using enumerative algorithms for the maximum stable set problem. Two approaches are used in the implementation of the enumerative algorithms for the maximum stable set problem, both using the bit parallel in the representation of the graph and on the operations with subsets of vertices. The first one is a known branch-and-bound algorithm and the second uses the Russian dolls method. In both cases, the use of bit parallel improve efficiency when the lower bounds are calculated based in a clique cover of the vertices. The results of computational experiments are presented as comparison between the two algorithms and as an assessment of the structures implemented. These results show that the algorithm based on the method Russian Dolls is more efficient regarding runtime and the memory consumed. Furthermore, the experimental results also show that the use stratification and hash tables also allow more efficiency in the case of sparse graphs.