Counting Background Knowledge Consistent Markov Equivalent Directed Acyclic Graphs

Vidya Sagar Sharma

Keywords: MEC, AMO, PEO, FPT, Clique, Clique Tree, Markov Equivalence Class, Acyclic Moral Orientation, Perfect Elimination Ordering, Causality, Causal Graph, Computational Complexity, Algorithm, Topological Ordering, Fixed Parameter Tractable

TL;DR: This paper gives a fixed-parameter tractable algorithm to count the number of directed acyclic graphs in a Markov equivalence class under background knowledge constraint.

Abstract: We study the problem of counting the number of directed acyclic graphs in a Markov equivalence class (MEC) that are consistent with background knowledge specified in the form of the directions of some additional edges in the MEC. A polynomial-time algorithm for the special case of the problem when no background knowledge constraints are specified was given by Wienöbst, Bannach, and Liśkiewicz (AAAI 2021), who also showed that the general case is NP-hard (in fact, #P-hard). In this paper, we show that the problem is nevertheless tractable in an interesting class of instances, by establishing that it is ``fixed-parameter tractable'': we give an algorithm that runs in time $O(k! k^2 n^4)$, where $n$ is the number of nodes in the MEC and $k$ is the maximum number of nodes in any maximal clique of the MEC that participate in the specified background knowledge constraints. In particular, our algorithm run in polynomial time in the well-studied special case of MECs of bounded treewidth or bounded maximum clique size.