Empirical hardness of finding optimal Bayesian network structures: algorithm selection and runtime prediction

被引:23
作者
Malone, Brandon [1 ]
Kangas, Kustaa [2 ]
Jarvisalo, Matti [2 ]
Koivisto, Mikko [2 ]
Myllymaki, Petri [2 ]
机构
[1] NEC Labs Europe, Heidelberg, Germany
[2] Univ Helsinki, HIIT, Dept Comp Sci, Helsinki, Finland
基金
芬兰科学院;
关键词
Bayesian networks; Structure learning; Algorithm selection; Hyperparameter optimization; Empirical hardness; Algorithm portfolio; Runtime prediction; STRUCTURE DISCOVERY; TREEWIDTH;
D O I
10.1007/s10994-017-5680-2
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Various algorithms have been proposed for finding a Bayesian network structure that is guaranteed to maximize a given scoring function. Implementations of state-of-the-art algorithms, solvers, for this Bayesian network structure learning problem rely on adaptive search strategies, such as branch-and-bound and integer linear programming techniques. Thus, the time requirements of the solvers are not well characterized by simple functions of the instance size. Furthermore, no single solver dominates the others in speed. Given a problem instance, it is thus a priori unclear which solver will perform best and how fast it will solve the instance. We show that for a given solver the hardness of a problem instance can be efficiently predicted based on a collection of non-trivial features which go beyond the basic parameters of instance size. Specifically, we train and test statistical models on empirical data, based on the largest evaluation of state-of-the-art exact solvers to date. We demonstrate that we can predict the runtimes to a reasonable degree of accuracy. These predictions enable effective selection of solvers that perform well in terms of runtimes on a particular instance. Thus, this work contributes a highly efficient portfolio solver that makes use of several individual solvers.
引用
收藏
页码:247 / 283
页数:37
相关论文
共 72 条
  • [1] SCIP: solving constraint integer programs
    Achterberg, Tobias
    [J]. MATHEMATICAL PROGRAMMING COMPUTATION, 2009, 1 (01) : 1 - 41
  • [2] [Anonymous], 2009, SIGKDD Explorations, DOI DOI 10.1145/1656274.1656278
  • [3] [Anonymous], 2007, P INT C ART INT STAT
  • [4] [Anonymous], 2006, P 22 C UNCERTAINTY A
  • [5] [Anonymous], 2014, ACM SIGKDD Explor. Newsl., DOI [DOI 10.1145/2641190.2641198, 10.1145/2641190.2641198]
  • [6] [Anonymous], 2011, P C UNCERTAINTY ARTI
  • [7] Overview and analysis of the SAT Challenge 2012 solver competition
    Balint, Adrian
    Belov, Anton
    Jarvisalo, Matti
    Sinz, Carsten
    [J]. ARTIFICIAL INTELLIGENCE, 2015, 223 : 120 - 155
  • [8] Barlett M., 2013, P C UNC ART INT, P182
  • [9] Bartlett M., 2015, Artificial Intelligence, V244, P258, DOI [DOI 10.1016/J.ARTINT.2015.03.003, 10.1016/j.artint.2015.03.003, 10.1016/J.ARTINT.2015.03.003]
  • [10] Berg J, 2014, JMLR WORKSH CONF PRO, V33, P86