Causal Inference with Conditional Instruments Using Deep Generative Models

Authors: Debo Cheng, Ziqi Xu, Jiuyong Li, Lin Liu, Jixue Liu, Thuc Duy Le

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on synthetic and real-world datasets show that our method outperforms the existing IV methods. Extensive experiments on a wide range of synthetic and real-world datasets show that the causal effects estimated using the CIVs and conditioning sets obtained by CIV.VAE have the smallest estimation error compared with the state-of-the-art causal effect estimators.
Researcher Affiliation Academia 1 School of Computer Science and Engineering, Guangxi Normal University, Guilin, 541004, China 2 Uni SA STEM, University of South Australia, Adelaide, SA, 5095, Australia
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks. It describes the model architecture and mathematical formulations but no step-by-step algorithm.
Open Source Code No The paper mentions that code for *other* methods (Deep IV, CEVAE, TEDVAE) is available on GitHub or through Python libraries, but it does not provide an explicit link or statement about open-sourcing the code for its *own* proposed method, CIV.VAE. For example: "The program of Deep IV is retrieved from the authors Git Hub 2. The implementation of CEVAE is obtained from the Python library pyro (Bingham, Chen et al. 2019) and the implementation of TEDVAE is downloaded from the authors Git Hub 4."
Open Datasets Yes For the second part, we use three real-world datasets, Schoolingreturns (Card 1993), 401k (Wooldridge 2010) and Sachs (Sachs, Perez et al. 2005), which have reference causal effect values available in literature.
Dataset Splits No The paper does not specify exact training, validation, and test dataset splits (e.g., percentages or sample counts). It mentions generating synthetic datasets of various sizes and using existing real-world datasets, but no specific splitting strategy for reproducibility. For synthetic data, it says "30 synthetic datasets for each sample size are generated", indicating multiple runs but not a train/validation/test split for a single dataset.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU models, CPU models, memory specifications) used for running the experiments. It only mentions the software libraries used.
Software Dependencies No The paper mentions software libraries used: "We use Python and the libraries including pytorch (Paszke, Gross et al. 2019), pyro (Bingham, Chen et al. 2019) and scikit-learn (Pedregosa et al. 2011) to implement our CIV.VAE method." However, it does not provide specific version numbers for these libraries, which is required for a reproducible description of ancillary software.
Experiment Setup No The paper states: "We provide the details of our CIV.VAE implementation and the parameters setting in the supplement." This indicates that specific experimental setup details, such as hyperparameters, are not included in the main text.