An Eigenmodel for Dynamic Multilayer Networks

Authors: Joshua Daniel Loyal, Yuguo Chen

JMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the estimation procedure s accuracy and scalability on simulated networks. We apply the model to two real-world problems: discerning regional conflicts in a data set of international relations and quantifying infectious disease spread throughout a school based on the student s daily contact patterns.
Researcher Affiliation Academia Joshua Daniel Loyal EMAIL Department of Statistics Florida State University Tallahassee, FL 32308, USA Yuguo Chen EMAIL Department of Statistics University of Illinois at Urbana-Champaign Champaign, IL 61820, USA
Pseudocode Yes Algorithm 1: Coordinate ascent variational inference for the eigenmodel for dynamic multilayer networks. Appendix C contains the details of Algorithms 2 5.
Open Source Code Yes A repository for the replication code is available on Github (Loyal, 2023).
Open Datasets Yes The raw data consists of (source actor, target actor, event type, time-stamp) tuples collected by the Integrated Crisis Early Warning System (ICEWS) project (Boschee et al., 2015)... The contact networks were collected by the Socio Patterns collaboration (http://www.sociopatterns.org) and initially analyzed in Stehl e et al. (2011).
Dataset Splits Yes Specifically, we randomly labeled 20% of the dyads as held-out data and removed them during estimation.
Hardware Specification Yes The machine used for these benchmarks was a 2021 Mac Book Pro with an Apple M1 Pro processor and a memory of 32 GB.
Software Dependencies No The paper does not explicitly provide specific software dependencies with version numbers used for the implementation or experiments.
Experiment Setup Yes We set ϵ = 0.05 and E [τ 2 δ ] = 10. For σ2 δ, we set cσ2 δ = 2 and dσ2 δ = 2. For the latent position s variational distributions, we set the variances and cross-covariances to the identity matrix Id. We chose uninformative priors for the inverse-Wishart distribution by setting ν = 2 + d and V = Id. For σ2, we set cσ2 = 2 and dσ2 = 2. Lastly, for the reference layer, we set ρ = 1/2 and Σλk = 10Id. In addition, we set the prior variance to σ2 λ = 4. Lastly, we initialized the mean of the P olya-gamma random variables, µωk ijt, to zero.