Robust Recursive Partitioning for Heterogeneous Treatment Effects with Uncertainty Quantification

Authors: Hyun-Suk Lee, Yao Zhang, William Zame, Cong Shen, Jang-Won Lee, Mihaela van der Schaar

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments using synthetic and semi-synthetic datasets (based on real-world data) demonstrate that R2P outperforms state-of-the-art methods by more robustly identifying subgroups while providing much narrower confidence intervals.
Researcher Affiliation Academia Hyun-Suk Lee Sejong University hyunsuk@sejong.ac.kr Yao Zhang University of Cambridge yz555@cam.ac.uk William R. Zame UCLA zame@econ.ucla.edu Cong Shen University of Virginia cong@virginia.edu Jang-Won Lee Yonsei University jangwon@yonsei.ac.kr Mihaela van der Schaar University of Cambridge UCLA The Alan Turing Institute mv472@cam.ac.uk
Pseudocode Yes Algorithm 1 Robust Recursive Partitioning
Open Source Code Yes The code of R2P is available at: https://bitbucket.org/mvdschaar/mlforhealthlabpub.
Open Datasets Yes The two semi-synthetic datasets are based on real world data; the first uses the Infant Health and Development Program (IHDP) dataset [24] and the second uses the Collaborative Perinatal Project (CPP) dataset [25].
Dataset Splits No The paper describes internal splitting for the SCR method (training set I1 and validation set I2) and for subgroup partitioning, but does not provide specific train/validation/test dataset splits for the overall experimental evaluation on the named datasets.
Hardware Specification No The paper does not provide any specific hardware specifications (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software components and methods like 'causal multi-task Gaussian process (CMGP)', 'Random Forest', 'multi-task Gaussian processes', and 'deep neural networks', but does not provide specific version numbers for any software or libraries used in the experiments.
Experiment Setup Yes We set the miscoverage rate to be α = 0.05, so we demand a 95% ITE coverage rate.