论文部分内容阅读
针对基于约束方法学习贝叶斯网络(BN)结构的不足,以及随着条件集的增大,利用统计方法进行条件独立(CI)测试不稳定等问题,提出一种基于最大主子图分解(MPD)的BN等价类学习算法.该算法首先通过MPD分解技术对BN的道德图进行分解;然后利用0阶和1阶CI测试识别部分子图中的V结构,对于初步未定的V结构利用局部评分搜索确定,从而避免了冗余检验,有效地减小了条件集的维数,并且提高了算法的效率.理论证明和实验结果均表明了所提出算法的有效性和合理性.
In order to solve the problem of learning BN structure based on constrained method and instability of conditional independence (CI) test with statistic method, a method based on MPD ) Algorithm of BN equivalence class learning algorithm, which decomposes the ethogram of BN by MPD decomposition technique first, and then uses the 0th order and the first order CI test to identify the V structure in some subgraphs, Which can avoid the redundancy test, effectively reduce the dimension of the condition set and improve the efficiency of the algorithm.The theoretical proof and experimental results show the effectiveness and rationality of the proposed algorithm.