Dependent Relational Gamma Process Models for Longitudinal Networks

Authors: Sikun Yang, Heinz Koeppl

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on a simulation study and three realworld temporal network data sets demonstrate the model s capability, competitive performance and scalability compared to state-of-the-art methods.
Researcher Affiliation Academia Sikun Yang 1 Heinz Koeppl 1 1Department of Electrical Engineering and Information Technology, Technische Universit at Darmstadt, Germany.
Pseudocode No The paper states: 'The complete Gibbs sampling algorithm and additional experimental results are presented in the supplementary material.' There is no pseudocode or algorithm block directly in the main body of the paper.
Open Source Code No The paper provides links to the code for baseline methods (DRIFT, DSBM, HGPEPM) but does not provide a link or explicit statement about the availability of the source code for the proposed DRGPM model.
Open Datasets Yes We consider the following data sets: (1) Face-to-face dynamic contacts network (FFDC): This dataset (Mastrandrea et al., 2015) records timestamped face-to-face contacts among 180 students for 7 school days. (2) DBLP: The DBLP co-authorship network data (Asur et al., 2009) contains the co-authorship information among 958 authors over ten years (1997-2006)... (3) Enron: The Enron data4 contains 517,431 emails among 151 users over 38 months (from May 1999 to June 2002). 4https://www.cs.cmu.edu/ enron/.
Dataset Splits No We randomly hold out 20% of the network entries (either links or non-links) for each snapshot as test data, and use the remaining 80% to predict the held-out entries. No explicit mention of a separate validation split percentage or count was found.
Hardware Specification Yes All the experiments were run on a standard desktop with 2.7 GHz CPU and 24 GB RAM.
Software Dependencies No The paper states: '(all models are implemented in Matlab)'. However, no specific version numbers for Matlab or any other software dependencies are provided.
Experiment Setup Yes In the experiments, we set the hyperparameters for our model as γ0 = 1, β = 1, c = 1, τ = 1. Unless otherwise stated, we use K = N/2 for initilization, where N is the number of nodes. For all probabilistic methods, we use 2000 burn-in iterations, and collect 1000 samples from the model posterior distribution.