CAIM(Class-Attribute InterdependenceMaximization) is one of the stateof-the-art algorithms for discretizing data for which classes are known. However, itmay take a long time when run on high-dimensional large-scale data, with large numberof attributes and/or instances. This paper presents a solution to this problem byintroducing a GPU-based implementation of the CAIM algorithm that significantlyspeeds up the discretization process on big complex data sets. The GPU-based implementationis scalable to multiple GPU devices and enables the use of concurrentkernels execution capabilities ofmodernGPUs. The CAIMGPU-basedmodel is evaluatedand compared with the original CAIM using single and multi-threaded parallelconfigurations on 40 data sets with different characteristics. The results show greatspeedup, up to 139 times faster using 4 GPUs, which makes discretization of bigdata efficient and manageable. For example, discretization time of one big data set isreduced from 2 hours to less than 2 minutes
展开▼
