Variational Inference for sparse network reconstruction from count data

Authors: Julien Chiquet, Stephane Robin, Mahendra Mariadassou

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We introduce the model in Section 2, the variational inference procedure in Section 3, simulation results in Section 4 and a reanalysis of two datasets in Section 5
Researcher Affiliation Academia 1MIA 518, Agro Paritech/INRA, Universit e Paris-Saclay, Paris, France 2Ma IAGE, INRA, Universit e Paris-Saclay, Jouyen-Josas, France.
Pseudocode No The paper describes the 'Inference Algorithm' in Section 3.2 but presents it in paragraph form with mathematical expressions rather than structured pseudocode or a clearly labeled algorithm block.
Open Source Code Yes We implemented our algorithm in a R/C++ package PLNmodels, available on github https: //github.com/jchiquet/PLNmodels.
Open Datasets Yes Data were downloaded from the French open data platform data.gouv.fr and filtered to remove stations with no votes.
Dataset Splits No The paper describes using resampling and subsampling (e.g., 'B = 50 subsamples of size m = 10 n') for model selection (StARS) and mentions varying sample size 'n' in simulations. However, it does not specify explicit training, validation, and test dataset splits with percentages or sample counts for model reproduction.
Hardware Specification No The paper does not provide any specific hardware details such as GPU or CPU models, processor types, or memory used for running the experiments.
Software Dependencies No The paper mentions software like the 'PLNmodels R/C++ package', 'nlopt library', 'glassofast R package', and 'RNAseq Net R-package', but it does not specify version numbers for any of them.
Experiment Setup Yes We fix the number of variables to p = 50 in all our experiments. The sampling efforts Ni are drawn from a negative binomial distribution with mean 1000 and variance 1000 + 10002/ν. We always set v = 0.3, u = 0.1 in our simulations. We use B = 50 subsamples of size m = 10 n and 2β = 0.05 as default, as suggested in Liu et al. Here, we replace loglik in EBIC by its variational surrogate J(Y; ˆΩ) and use γ = 0.5 as recommended by Foygel & Drton for GGM.