The impact of variable ordering on Bayesian network structure learning

被引:4
作者
Kitson, Neville K. [1 ]
Constantinou, Anthony C. [1 ]
机构
[1] Queen Mary Univ London, Machine Intelligence & Decis Syst MInDS Grp, Mile End Rd, London E1 4NS, England
关键词
Bayesian networks; Directed acyclic graphs; Probabilistic graphical models; Structure learning; DISCOVERY; INDUCTION;
D O I
10.1007/s10618-024-01044-9
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Causal Bayesian Networks (CBNs) provide an important tool for reasoning under uncertainty with potential application to many complex causal systems. Structure learning algorithms that can tell us something about the causal structure of these systems are becoming increasingly important. In the literature, the validity of these algorithms is often tested for sensitivity over varying sample sizes, hyper-parameters, and occasionally objective functions, but the effect of the order in which the variables are read from data is rarely quantified. We show that many commonly-used algorithms, both established and state-of-the-art, are more sensitive to variable ordering than these other factors when learning CBNs from discrete variables. This effect is strongest in hill-climbing and its variants where we explain how it arises, but extends to hybrid, and to a lesser-extent, constraint-based algorithms. Because the variable ordering is arbitrary, any significant effect it has on learnt graph accuracy is concerning, and raises questions about the validity of both many older and more recent results produced by these algorithms in practical applications and their rankings in performance evaluations.
引用
收藏
页码:2545 / 2569
页数:25
相关论文
共 47 条
  • [1] [Anonymous], 1977, Combinatorial Mathematics V, DOI [10.1007/BFb0069178, DOI 10.1007/BFB0069178]
  • [2] [Anonymous], 2009, Probabilistic graphical models: principles and techniques
  • [3] Integer Linear Programming for the Bayesian network structure learning problem
    Bartlett, Mark
    Cussens, James
    [J]. ARTIFICIAL INTELLIGENCE, 2017, 244 : 258 - 271
  • [4] Improved K2 algorithm for Bayesian network structure learning
    Behjati, Shahab
    Beigy, Hamid
    [J]. ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE, 2020, 91 (91)
  • [5] Bernstein DI, 2020, PR MACH LEARN RES, V108, P4098
  • [6] BOUCKAERT RR, 1992, UNCERTAINTY IN ARTIFICIAL INTELLIGENCE, P9
  • [7] Bouckaert RR, 1994, P 10 C UNC ART INT S, P102
  • [8] On inclusion-driven learning of Bayesian networks
    Castelo, R
    Kocka, T
    [J]. JOURNAL OF MACHINE LEARNING RESEARCH, 2004, 4 (04) : 527 - 574
  • [9] Chickering D. M., 2003, Journal of Machine Learning Research, V3, P507, DOI 10.1162/153244303321897717
  • [10] Colombo D, 2014, J MACH LEARN RES, V15, P3741