Shlukování pomocí reprezentantů pro velká data

Abstract

The aim of this study is realization of selected clustering algorithm and subsequent adjustment for big data processing. In this study CLARANS algorithm was implemented in form suitable for parallel processing on a graphic card using the CUDA architecture. The study presents theoretical foundations regarding clustering, including solving difficulties related to the processing of big data. Above the proposed algorithm implementation a number of experiments are performed. Within these experiments the algorithm is compared with the k-medoids algorithm within the algorithm running times and the quality of the founded clustering. Experiments include testing the newly designed approach for selecting new neighbors in the algorithm and testing the ability of algorithm to deal with big data.