Acceleration of Feature Subset Selection Using CUDA
Jun Yang, Siyuan Jing · 2018
Rough sets have been proven to be an effective tool for feature subset selection, which is a key step in various machine learning tasks. However, this task is very time consuming. To address this problem, graphics processing unit (GPU), which is a popular device of high performance computing, is applied to accelerate a sorting-based algorithm of feature subset selection. The proposed algorithm is well designed by CUDA programming framework. To obtain great performance gain, two critical steps in rough sets based feature subset selection, which are computation of equivalence class and feature significance, are both executed on GPU. Experimental results show that the proposed algorithm is efficient and it can scale well on large data sets.