Conditional independence testing under misspecified inductive biases

Authors: Felipe Maia Polo, Yuekai Sun, Moulinath Banerjee

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct experiments with artificial and real data, showcasing the usefulness of our theory and methods.
Researcher Affiliation Academia Felipe Maia Polo Department of Statistics University of Michigan felipemaiapolo@gmail.com Yuekai Sun Department of Statistics University of Michigan yuekai@umich.edu Moulinath Banerjee Department of Statistics University of Michigan moulib@umich.edu
Pseudocode Yes Algorithm 1: Obtaining p-value for the RBPT
Open Source Code Yes Code in https://github.com/felipemaiapolo/cit.
Open Datasets Yes For our subsequent experiments, we employ the car insurance dataset examined by Angwin et al. [2].
Dataset Splits Yes the training (resp. test) dataset consists of 800 (resp. 200) entries
Hardware Specification Yes all in a Mac Book Air 2020 M1.
Software Dependencies No The paper mentions software like "Python script" and "CatBoost regressor" but does not specify their version numbers or other library versions.
Experiment Setup Yes We assume α = 10% and ℓ(ˆy, y) = (ˆy y)2. ... every predictor we employ operates on linear regression. ... RESIT employs Spearman s correlation between residuals as a test statistic ... For RBPT2, we use a Cat Boost regressor [26] to yield the Rao-Blackwellized predictor. We resort to logistic regression for estimating the distribution of X | Z used by RBPT, GCM, CRT, and CPT.