Bayesian Data Sketching for Varying Coefficient Regression Models

Authors: Rajarshi Guhaniyogi, Laura Baracaldo, Sudipto Banerjee

JMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We use simulation experiments and analyze remote sensed vegetation data to empirically illustrate the inferential and computational efficiency of our approach. Keywords: B-splines, Predictive Process, Posterior contraction, Random compression matrix, Varying coefficient models. ... Section 4 demonstrates performance of the proposed approach with simulation examples and a forestry data analysis.
Researcher Affiliation Academia Rajarshi Guhaniyogi EMAIL Department of Statistics Texas A & M University College Station, TX 77843-3143, USA. Laura Baracaldo EMAIL Department of Statistics and Applied Probability University of California Santa Barbara Santa Barbara, CA 93106-3110, USA. Sudipto Banerjee EMAIL UCLA Department of Biostatistics University of California Los Angeles Los Angeles, CA 90095-1772, USA.
Pseudocode Yes Algorithm 1: Parametric Inference from the Proposed Model ... Algorithm 2: Predictive Inference from the Proposed Model
Open Source Code Yes Source codes for these experiments are available from https://github.com/ Laura Baracaldo/Bayesian-Data-Sketching-in-Spatial-Regression-Models.
Open Datasets Yes We implement geo S to analyze vegetation data gathered through the Moderate Resolution Imaging Spectroradiometer (MODIS), which resides aboard the Terra and Aqua platforms on NASA spacecrafts. ... The data, which were downloaded using the R package MODIS, comprises 133, 000 observed locations where the response was measured using the MODIS tool over a 16-day period in April 2016.
Dataset Splits Yes We retained N = 113, 000 observations (randomly chosen) for model fitting and used the rest for prediction.
Hardware Specification Yes We implemented our models in the R statistical computing environment on a Dell XPS 13 PC with Intel Core i7-8550U CPU @ 4.00GHz processors at 16 GB of RAM.
Software Dependencies No The paper mentions using 'R statistical computing environment' and 'mcmcse package in R' and 'sp Bayes package in R' but does not specify version numbers for any of these.
Experiment Setup Yes We complete the hierarchical specification by assigning independent IG(2, 0.1) priors (mean 0.1 with infinite variance) for σ2 and τ 2j for each j = 1, . . . , P. ... For each of our simulation data sets we ran a single-threaded MCMC chain for 5000 iterations. Posterior inference was based upon 2000 samples retained after adequate convergence was diagnosed using Monte Carlo standard errors and effective sample sizes (ESS) using the mcmcse package in R.