Conditional Instrumental Variable Regression with Representation Learning for Causal Inference
Authors: Debo Cheng, Ziqi Xu, Jiuyong Li, Lin Liu, Jixue Liu, Thuc Duy Le
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on synthetic and two real-world datasets show the competitive performance of CBRL.CIV against state-of-the-art IV-based estimators and superiority in dealing with the non-linear situation. |
| Researcher Affiliation | Collaboration | Debo Cheng 1 , Ziqi Xu 1,2 , Jiuyong Li 1, Lin Liu 1, Jixue Liu 1 & Thuc Duy Le 1 University of South Australia 1 Data61, CSIRO 2 {firstname.lastname}@unisa.edu.au 1, ziqi.xu@data61.csiro.au 2 |
| Pseudocode | Yes | The pseudo-code of the proposed CBRL.CIV method is provided in Appendix B.1. |
| Open Source Code | Yes | The code of CBRL.CIV, network parameters, and parameter tuning are provided in the supplementary materials. |
| Open Datasets | Yes | We also conduct experiments on two real-world datasets, IHDP (Hill, 2011) and Twins (Shalit et al., 2017), to evaluate the performance of CBRL.CIV method and demonstrate its potential in realworld applications. |
| Dataset Splits | Yes | The experiments on the two real-world datasets are also conducted over 30 replications with the 63/27/10 splitting of train/validation/test samples. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions software tools like 'Python programming language', 'multi-layer perceptrons', 'SGD', and 'ADAM method', but does not provide specific version numbers for these software components or libraries. |
| Experiment Setup | Yes | We used multi-layer perceptrons with Re LU activation function and Batch Norm to implement the logistic regression networks for CIV regression and treatment regression. For optimising the CIV and treatment regression networks, we used stochastic gradient descent (SGD) (Duchi et al., 2011). Furthermore, we used the ADAM method (Kingma & Ba, 2014) to optimise the outcome network jointing the confounding balance modules. To prevent overfitting, we employed an l2-regularisation term to regularise the objective function in Eq. (6). The code of CBRL.CIV, network parameters, and parameter tuning are provided in the supplementary materials. The implementations and parameter settings of compared estimators are provided in Appendix C.1. ... For the two tuning parameters α and β, we select them from the range {0.0001, 0.001, 0.01, 0.1, 1}. |