BAST: Bayesian Additive Regression Spanning Trees for Complex Constrained Domain

Authors: Zhao Tang Luo, Huiyan Sang, Bani Mallick

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate the model performance by simulation experiments and a real chlorophyll data set in Aral Sea.
Researcher Affiliation Academia Zhao Tang Luo Department of Statistics Texas A&M University ztluo@stat.tamu.edu Huiyan Sang Department of Statistics Texas A&M University huiyan@stat.tamu.edu Bani Mallick Department of Statistics Texas A&M University bmallick@stat.tamu.edu
Pseudocode No The paper describes algorithms and steps for Bayesian inference in prose, but it does not present any structured pseudocode or algorithm blocks.
Open Source Code Yes The code of BAST is available at https://github.com/ztluostat/BAST.
Open Datasets Yes We apply BAST to analyze average remote sensed chlorophyll data in the Aral data over 1998-2002, which are available in the R package gamair [42].
Dataset Splits Yes We first compare the prediction performance of all the models via 10-fold cross-validation.
Hardware Specification Yes All experiments are conducted on a single CPU core (Intel Xeon E5-2630 v4 CPU @ 2.20GHz) with 10GB of memory.
Software Dependencies No The paper mentions using the "R package gamair [42]" for a dataset, but it does not provide specific version numbers for the primary software components used to implement and run BAST (e.g., Python, PyTorch, TensorFlow, specific libraries).
Experiment Setup Yes We use M = 20 weak learners and set λk = 4 and k = 10 to restrict the size of each partition. ... The probabilities for MCMC moves are set as rb = rd = rc = 0.3 and rh = 0.1, with adjustments for cases where km = 1 or k. We run the MCMC for both BAST and BART for 20, 000 iterations, discarding the first half and retaining samples every 5 iterations.