Efficient Online Estimation of Causal Effects by Deciding What to Observe

Authors: Shantanu Gupta, Zachary Lipton, David Childers

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we validate our findings experimentally 1 (Section 6). We validate our methods on synthetic data generated from known causal graphs (see Appendix D for parameter values and moment conditions used).
Researcher Affiliation Academia Shantanu Gupta, Zachary C. Lipton, David Childers Carnegie Mellon University {shantang,zlipton,dchilders}@cmu.edu
Pseudocode Yes Figure 1: Algorithms for OMS-ETC and OMS-ETG.
Open Source Code Yes The code and data are available at https://www.github.com/acmi-lab/ online-moment-selection.
Open Datasets Yes Finally, we demonstrate the effectiveness of our methods on two semi-synthetic datasets: the Infant Health Development Program (IHDP) dataset [19] and a Vietnam era draft lottery dataset [2]. Replication data for: Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence from Social Security Administrative Records, 2009. URL https://doi.org/10. 7910/DVN/PLF0YL.
Dataset Splits No No specific train/validation/test dataset splits (e.g., percentages, sample counts, or explicit standard splits) are mentioned. The paper describes generating semi-synthetic data and states 'The MSE values are computed across 12, 000 runs.'
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, processor types, memory amounts) used for running experiments are mentioned in the paper.
Software Dependencies No No specific software dependencies with version numbers (e.g., library names like 'PyTorch 1.9' or solver versions like 'CPLEX 12.4') are provided. The paper only mentions using a 'regularized GMM weight matrix' and implies Python use given the GitHub link.
Experiment Setup No The paper mentions 'For OMS-ETG, we use the regularized GMM weight matrix with λW := 0.01' and 'The MSE values are computed across 12, 000 runs'. However, it does not provide specific hyperparameter values for model training (e.g., learning rates, batch sizes, epochs) or detailed system-level configurations.