Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Restrictions of PCBNs for integration-free computations

Alexis Derumigny, Niels Horsman, Dorota Kurowicka

TL;DR

This paper addresses the computational bottleneck in PCBNs where density evaluation and simulation may require high-dimensional integration. It provides a complete graph-level characterization: integration-free computations are possible if and only if the DAG contains no active cycles or interfering v-structures, and offers an algorithm to construct parental orders that guarantee integration-free costs. The authors also establish the asymptotic normality of stepwise estimators for PCBNs under these restricted conditions and demonstrate favorable finite-sample behavior in simulations. The proposed approach enables scalable, flexible modeling with copula-based dependencies while controlling computational complexity, with estimation implemented via a stepwise rank-based estimating-equations framework. Overall, the work yields practical guidelines and theoretical guarantees for integrating PCBNs in high-dimensional settings.

Abstract

The pair-copula Bayesian Networks (PCBN) are graphical models composed of a directed acyclic graph (DAG) that represents (conditional) independence in a joint distribution. The nodes of the DAG are associated with marginal densities, and arcs are assigned with bivariate (conditional) copulas following a prescribed collection of parental orders. The choice of marginal densities and copulas is unconstrained. However, the simulation and inference of a PCBN model may necessitate possibly high-dimensional integration. We present the full characterization of DAGs that do not require any integration for density evaluation or simulations. Furthermore, we propose an algorithm that can find all possible parental orders that do not lead to (expensive) integration. Finally, we show the asymptotic normality of estimators of PCBN models using stepwise estimating equations. Such estimators can be computed effectively if the PCBN does not require integration. A simulation study shows the good finite-sample properties of our estimators.

Restrictions of PCBNs for integration-free computations

TL;DR

This paper addresses the computational bottleneck in PCBNs where density evaluation and simulation may require high-dimensional integration. It provides a complete graph-level characterization: integration-free computations are possible if and only if the DAG contains no active cycles or interfering v-structures, and offers an algorithm to construct parental orders that guarantee integration-free costs. The authors also establish the asymptotic normality of stepwise estimators for PCBNs under these restricted conditions and demonstrate favorable finite-sample behavior in simulations. The proposed approach enables scalable, flexible modeling with copula-based dependencies while controlling computational complexity, with estimation implemented via a stepwise rank-based estimating-equations framework. Overall, the work yields practical guidelines and theoretical guarantees for integrating PCBNs in high-dimensional settings.

Abstract

The pair-copula Bayesian Networks (PCBN) are graphical models composed of a directed acyclic graph (DAG) that represents (conditional) independence in a joint distribution. The nodes of the DAG are associated with marginal densities, and arcs are assigned with bivariate (conditional) copulas following a prescribed collection of parental orders. The choice of marginal densities and copulas is unconstrained. However, the simulation and inference of a PCBN model may necessitate possibly high-dimensional integration. We present the full characterization of DAGs that do not require any integration for density evaluation or simulations. Furthermore, we propose an algorithm that can find all possible parental orders that do not lead to (expensive) integration. Finally, we show the asymptotic normality of estimators of PCBN models using stepwise estimating equations. Such estimators can be computed effectively if the PCBN does not require integration. A simulation study shows the good finite-sample properties of our estimators.

Paper Structure

This paper contains 40 sections, 38 theorems, 111 equations, 19 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background on DAGs and BNs
  3. Directed graphs
  4. Bayesian networks
  5. Pair-Copula Bayesian Networks
  6. Introduction
  7. Problematic conditional margins for PCBNs
  8. Active cycles
  9. Interfering v-structures
  10. Restricted PCBNs
  11. Possible candidates and algorithm
  12. B-sets
  13. Explicit characterization of the set of possible candidates
  14. Example
  15. Proof of Theorem \ref{['thm:existence_possible_candidates']}
  16. ...and 25 more sections

Key Result

Lemma 3.1

The joint copula density $c_{ \mathbf{X} }({ \mathbf{u} })$ (and therefore $f_{ \mathbf{X} }({ \mathbf{x} })$) of a PCBN can be computed without integration if and only if

Figures (19)

  • Figure 1: Diamond PCBN on four nodes.
  • Figure 2: PCBN where node $X_4$ has corresponding parental order $1 <_4 2 <_4 3$.
  • Figure 3: PCBN where $C_{2,4|1,5} = C_{2,4|1}$ is specified.
  • Figure 4: PCBNs where the computation of $u_{3|2}$ can be done without integrating for (b) and (c), but not for (a).
  • Figure 5: PCBN with interfering v-structures.
  • ...and 14 more figures

Theorems & Definitions (90)

  • Definition 2.1: d-separation
  • Definition 2.2: Bayesian network
  • Definition 3.1: Parental order
  • Definition 3.2: Pair-copula Bayesian network
  • Definition 3.3
  • Remark 3.1
  • Remark 3.2
  • Example 3.1
  • Example 3.2
  • Definition 3.4
  • ...and 80 more