HTILDE-RT: Um algoritmo para aprender árvores de Regressão Relacionais em grandes conjuntos de dados

Glauber M. C. MenezesGerson Zaverucha

Currently, modern organizations store their data under the form of relational databases which grow faster than hardware capacities. However, ex- tracting information from such databases has become crucial. In this work we propose HTILDE-RT, an algorithm to learn relational regression trees efficiently from huge databases. It is based on the ILP system TILDE and the propositi- onal system VFDT learner. The algorithm uses Hoeffding bound to scale up the learning process. We compared HTILDE-RT with TILDE-RT in two large datasets, each with two million examples, yielding more than three times faster learning times with no statistically significant difference in Pearson correlation coefficient.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: