Spike and slab variational Bayes for high dimensional logistic regression

Authors: Kolyan Ray, Botond Szabo, Gabriel Clara

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We confirm the improved performance of our VB algorithm over common sparse VB approaches in a numerical study. We verify in a numerical study that the empirical performance of the proposed method reflects these theoretical guarantees.
Researcher Affiliation Academia Kolyan Ray Department of Mathematics Imperial College London kolyan.ray@ic.ac.uk Botond Szabó Department of Mathematics Vrije Universiteit Amsterdam b.t.szabo@vu.nl Gabriel Clara Department of Mathematics Vrije Universiteit Amsterdam g.clara@student.vu.nl
Pseudocode Yes Algorithm 1: Modified CAVI for variational Bayes with Laplace slabs
Open Source Code Yes We are currently working on a more efficient implementation as an R-package sparsevb [15] that should improve the run-time. [15] CLARA, G., SZABO, B., AND RAY, K. sparsevb: spike and slab variational Bayes for linear and logistic regression, 2020. R package version 1.0.
Open Datasets No The paper describes synthetic data generation, e.g., "We take n = 250, p = 500 and X to be a standard Gaussian design matrix, i.e. Xij iid N(0, 1), and set the true signal θ0 = (2, 2, 0, . . . , 0)T to be s = 2 sparse." It does not provide access information for a publicly available or open dataset.
Dataset Splits No The paper mentions "n training examples" but does not specify any training, validation, or test dataset splits (e.g., percentages, counts, or a specific splitting methodology) needed for reproduction.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments were found in the paper.
Software Dependencies No The paper mentions "C++ using the Rcpp interface and used the Armadillo linear algebra library and ensmallen optimization library" but does not provide specific version numbers for these software components.
Experiment Setup Yes We take n = 250, p = 500 and X to be a standard Gaussian design matrix, i.e. Xij iid N(0, 1), and set the true signal θ0 = (2, 2, 0, . . . , 0)T to be s = 2 sparse. We ran the experiment 200 times for each method and report the means and standard deviations of the following performance measures:...