Empirical Gateaux Derivatives for Causal Inference

Authors: Michael Jordan, Yixin Wang, Angela Zhou

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Figure 1a we include an ( , λ)-plot for the simple case of AIPW (see [10] for more discussion and examples). We consider a one-dimensional case with uniformly distributed X, piecewiselinear Y , and smooth propensity scores that are logistic in sin(X). We use n = 500 and fix the bandwidth h = 0.05. Colors denote magnitude of the mean absolute error, included in text on the heatmap. Without loss of generality, we study the estimation of a mean under missingness, E[Y (1)]. Figure 1b illustrates the estimation error of various strategies with the comparable kernel-based estimates (DM is regression adjustment). We include further numerical experiments in the Appendix.
Researcher Affiliation Academia Michael I. Jordan Department of EECS and Department of Statistics University of California, Berkeley
Pseudocode Yes Algorithm 1 Empirical Gateaux derivatives
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets No The paper mentions generating data for a case study ('one-dimensional case with uniformly distributed X, piecewise-linear Y, and smooth propensity scores'), but it does not specify a publicly available dataset with a citation, link, or repository for access.
Dataset Splits No The paper describes experimental parameters such as 'n = 500' and a fixed bandwidth, but it does not provide specific details on training, validation, or test dataset splits or cross-validation setup.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, processors, or memory used for running the experiments.
Software Dependencies No The paper mentions 'scipy.optimize or r.optim packages' as general examples in a footnote, but it does not list specific software dependencies with version numbers that were used for its own experiments.
Experiment Setup Yes We use n = 500 and fix the bandwidth h = 0.05.