并行MapReduce PLS算法及其在光谱分析中的应用

Parallel PLS Aigorithm Using MapReduce and Its Aplication in Spectral Modeling

偏最小二乘(PLS)算法是常用的光谱建模算法,然而对于海量光谱处理情形,在单台计算机上建模及优化时间开销很大。基于MapReduce编程模式,提出了并行MapReduce PLS回归算法,包括并行数据标准化和并行主成分提取两个过程。在多台普通计算机上搭建Hadoop云计算集群平台,以近红外光谱处理为例,开展了算法验证实验。实验结果表明,基于MapReduce编程模式的并行PLS算法对海量近红外光谱数据集进行回归建模时,能有效提高建模速度,随计算机台数的增多可得到接近线性的加速比,并具有良好的扩展性。

Partial least squares（PLS） has been widely used in spectral analysis and modeling,and it is computation-intensive and time-demanding when dealing with massive data.To solve this problem effectively,a novel parallel PLS using MapReduce is proposed,which consists of two procedures,the parallelization of data standardizing and the parallelization of principal component computing.Using NIR spectral modeling as an example,experiments were conducted on a Hadoop cluster,which is a collection of ordinary computers.The experimental results demonstrate that the parallel PLS algorithm proposed can handle massive spectra,can significantly cut down the modeling time,and gains a basically linear speedup,and can be easily scaled up.