机构地区: 广东工业大学计算机学院
出 处: 《计算机应用与软件》 2016年第8期249-253,共5页
摘 要: 如何根据观察数据来推断因果网络结构是统计学和机器学习领域的重要问题。近年来学者们取得了许多研究成果,Li NGAM算法是其中一种经典的线性因果推断算法。但Li NGAM算法采用的剪枝策略时间复杂度较高,且在稀疏图上准确率低。为此,提出一种基于条件独立性测试的剪枝算法来解决这个问题。该算法首先将变量根据因果顺序重新排列,再按照该次序采用偏相关系数检验变量之间的条件独立性。大量的实验结果表明,基于条件独立性的剪枝算法在稀疏图上比Li NGAM的剪枝算法获得更高的准确率与执行效率。 H ow to con je cture causal netw ork structure according to the observed data is an im p o rta n t problem in the fie ld o f statistics andm achine le a rn in g . Q uite a few research achievem ents have been gained b y scholars in recent yea rs,am ong them L iN G A M a lg o rith m is a classicallin e a r causal infe rence a lg o rith m . H ow ever the p ru n in g p o lic y em ployed in L iN G A M a lg o rith m requires h ig h ru n tim e co m p le xity and p ro videspoor accuracy on sparse graphs. Therefore in th is pa per we present a novel p ru n in g m ethod to solve th is p ro b le m ,it is based on co n d itio n al independence te stin g . The a lg o rith m firs t rearranges the variables according to causal ord e r and the n em ploys p a rtia l co rre la tio n co e ffic ie n t toche ck the co n d itio n a l independence between variables according to new order. N um erous exp erim e ntal results in d ica te th a t the p ru n in g algorithm based on co n d itio n a l independence proposed in the pa per achieves h ig h e r accuracy w ith b e tte r ru n n in g tim e on sparse graph than the oneo f L iN G A M .
领 域: [自动化与计算机技术]