Causal Discovery from Heterogeneous/Nonstationary Data
Authors: Biwei Huang, Kun Zhang, Jiji Zhang, Joseph Ramsey, Ruben Sanchez-Romero, Clark Glymour, Bernhard Schölkopf
JMLR 2020 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on various synthetic and real-world data sets (task-fMRI and stock market data) are presented to demonstrate the efficacy of the proposed methods. |
| Researcher Affiliation | Academia | Department of Philosophy, Carnegie Mellon University, Pittsburgh, PA 15213. Department of Philosophy, Lingnan University, Tuen Mun, Hong Kong. Center for Molecular and Behavioral Neuroscience, Rutgers University, Newark, NJ 07102. Max Planck Institute for Intelligent Systems, Tübingen 72076, Germany. |
| Pseudocode | Yes | Algorithm 1 Detection of Changing Modules and Recovery of Causal Skeleton. Algorithm 2 Causal Direction Identification by Generalization of Invariance. Algorithm 3 Causal Direction Identification by Independent Changes of Causal Modules. Algorithm 4 KNV of Causal Modules P(Y |X, C). Algorithm 5 Detection of Changing Modules and Recovery of Causal Skeleton with both (Time-Varying) Instantaneous and Lagged Causal Relationships. |
| Open Source Code | No | The paper does not provide any explicit statements about the release of source code for the methodology described, nor does it include links to a code repository. |
| Open Datasets | Yes | We applied our methods to task f MRI data, which were recorded under a star/plus experiment (Wang et al., 2004). We then applied our methods to daily returns of stocks from Hong Kong (HK) and the United States (US), downloaded from Yahoo Finance. |
| Dataset Splits | No | The paper describes sample sizes and domain segmentation for synthetic data, and time periods for real-world data, but does not provide explicit training/test/validation dataset splits in the conventional machine learning sense for evaluating the method's performance on these datasets. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU or CPU models. |
| Software Dependencies | No | The paper mentions using PC for search and KCI test for conditional independence, and refers to specific Matlab code for a comparison method, but it does not provide specific version numbers for software dependencies for its own methodology. |
| Experiment Setup | Yes | For the KCI test, the significance level was 0.05. We used Gaussian kernels, and the hyperparameters, e.g. the kernel width, were set with empirical values; please refer to Zhang et al. (2011) for details. For illustration purposes, we used a single driving force for each module, which changes both causal strength b and noise parameter σ. We used Gaussian kernels both in kernel embedding of constructed joint distributions and kernel PCA. |