Causal Discovery via Conditional Independence Testing with Proxy Variables
Authors: Mingzhou Liu, Xinwei Sun, Yu Qiao, Yizhou Wang
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate the effectiveness of our procedure using both synthetic and real-world data. Code is publicly available at https://github.com/ lmz123321/proxy_causal_discovery. In this section, we apply our method to synthetic data and causal discovery on a real-world dataset. |
| Researcher Affiliation | Academia | 1School of Computer Science, Peking University (liumingzhou@stu.pku.edu.cn) 2Center on Frontiers of Computing Studies (CFCS), Peking University 3Sch. of Data Science, Fudan University 4Dep. of Automation, Shanghai Jiao Tong University 5Inst. for Artificial Intelligence, Peking University 6Nat l Eng. Research Center of Visual Technology, Peking University 7Nat l Key Lab of General Artificial Intelligence, Peking University. Correspondence to: Xinwei Sun <sunxinwei@fudan.edu.cn>. |
| Pseudocode | Yes | Algorithm 1 Discretizing W and X. |
| Open Source Code | Yes | Code is publicly available at https://github.com/ lmz123321/proxy_causal_discovery. |
| Open Datasets | Yes | Data extraction from MIMIC database. We consider the Medical Information Mart for Intensive Care (MIMIC III) database (Johnson et al., 2016), which contains the electronic health records for patients in the ICU. |
| Dataset Splits | No | The paper does not provide specific train/validation/test dataset splits. It describes synthetic data generation and data extraction from MIMIC-III, but not how these were partitioned into train/validation/test sets for model evaluation. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | We adopt the implementations provided in the causallearn and mliv packages, respectively. The paper mentions software packages but does not provide specific version numbers for them or other dependencies. |
| Experiment Setup | Yes | Implementation details. We set the significance level α to 0.05. We set the bin numbers of our method to l X = 14, l W = 12, respectively. For KCI and ICM, we adopt the implementations provided in the causallearn and mliv packages, respectively. For the procedure described by Miao, we implement the R code released in the paper and set l X = 3, l W = 2 by default. |