Structure Learning with Adaptive Random Neighborhood Informed MCMC

Authors: Xitong Liang, Alberto Caron, Samuel Livingstone, Jim Griffin

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental After introducing the technical novelties in PARNI-DAG, we empirically demonstrate its mixing efficiency and accuracy in learning DAG structures on a variety of experiments.1
Researcher Affiliation Academia Alberto Caron The Alan Turing Institute London, UK acaron@turing.ac.ukXitong Liang Department of Statistical Sciences University College London London, UK xitong.liang.18@ucl.ac.ukSamuel Livingstone Department of Statistical Sciences University College London London, UK samuel.livingstone@ucl.ac.ukJim Griffin Department of Statistical Sciences University College London London, UK j.griffin@ucl.ac.uk
Pseudocode Yes The full details and algorithmic pseudo code of PARNI-DAG are also provided in the supplementary material (Appendix D).
Open Source Code Yes 1Code to implement the PARNI-DAG proposal and replicate the experimental sections is available at https://github.com/Xitong Liang/The-PARNI-scheme/tree/main/Structure-Learning.
Open Datasets Yes We first consider the real-world protein-signalling dataset [Sachs et al., 2005], found also in Cundy et al. [2021], to test PARNI-DAG s mixing.
Dataset Splits No The paper mentions generating N=100 observations but does not provide specific details on train/validation/test splits, percentages, or explicit sample counts for the experimental setup.
Hardware Specification Yes Using Intel i7 2.80 GHz processor
Software Dependencies No The paper mentions 'R package bnlearn' but does not provide specific version numbers for it or any other software dependencies.
Experiment Setup Yes For ADR and PARNI-DAG, we use prior parameters g = 10 and h = 1/11.