分解合成系列快速計算法之平行化改進

Abstract

本計畫是將split and combine 系列的演算法推廣到平行計算，我們利用Python 程式的架構成功的將此系列的加速法推廣到多核心計算以及GPU 的計算。對於split and combine系列的計算法如何分群是一個重要的關鍵，我們經過這次的計畫補助理解了如何有效率的分群，讓分解合成計算更有效率。特別在資料維度高以及資料量大時，我們平行化的版本程式表現會更為優異。

We have implement the SCMDS method from the serial version to the multicore version. We use the python programming to do this work. Our project gives a very friendly introduction for parallel programming in python. In our experiments, we can see that the performance of multicore version of SCMDS makes the linear SCMDS better. When the data dimension is large and the size of data is huge, the performance of parallel version is pretty well. This parallel SCMDS is proved to be good for large data analysis.