计算机科学 ›› 2017, Vol. 44 ›› Issue (10): 71-74.doi: 10.11896/j.issn.1002-137X.2017.10.013
陈静,方建滨,唐滔,杨灿群
CHEN Jing, FANG Jian-bin, TANG Tao and YANG Can-qun
摘要: 用OpenCL语言标准设计并实现了推荐系统领域的两种经典算法:交替最小二乘法(Alternating Least Squares,ALS)与循环坐标下降法(Cyclic Coordinate Descent,CCD)。将其应用到CPU,GPU,MIC多核与众核平台上,探索了在该平台上影响算法性能的因子:潜在特征维数与线程个数。同时,将OpenCL实现的两种算法与CUDA和OpenMP的实现进行比较,得出了一系列结论。在同等条件下,与ALS算法相比,CCD算法的精度更高,收敛速度更快且更稳定,但所耗时间更长。ALS和CCD算法基于OpenCL的实现性能不亚于CUDA(CCD 上加速比为1.03x,ALS上加速比为1.2x)和OpenMP的实现(CCD与ALS上加速比大约为1.6~1.7x),并且两种算法在CPU平台上的性能均比GPU与MIC好。
[1] RODRIGUES A V,JORGE A,DUTRA I.Accelerating Recommender Systems using GPUs[C]∥ACM Symposium on Applied Computing.ACM,2015:879-884. [2] GATES M,ANZT H,KURZAK J,et al.Accelerating Collaborative Filtering Using Concepts from High Performance Computing[C]∥2015 IEEE International Conference on Big Data (Big Data).IEEE,2015:667-676. [3] PATEREK A.Improving regularized singular value decomposition for collaborative filtering[C]∥ACM International Con-ference on Knowledge Discovery and Data Mining.2007:39-42. [4] ZHOU Y H,WILKINSON D,SCHREIBER R,et al.Large-scale Parallel Collaborative Filtering for the Netflix Prize[C]∥Proceedings of the 4th international conference on Algorithmic Aspects in Information and Management.2008:337-348. [5] YU H F,HSIEH C J,SI S,et al.Scalable Coordinate Descent Approaches to Parallel Matrix Factorization for Recommender Systems[C]∥2013 IEEE 13th International Conference on Data Mining(2012).2012:765-774. [6] KOREN Y,BELL R,VOLINSKY C.Matrix Factorization Tech-niques for Recommender Systems[J].Computer,2009,2(8):30-37. [7] ZHUANG Y,CHIN W S,JUAN W C,et al.A Fast Parallel SGD for Matrix Factorization in Shared Memory Systems[C]∥Proceedings of ACM Recommender Systems 2013.2013:249-256. |
No related articles found! |
|