Testing Independence Between Linear Combinations for Causal Discovery

Authors: Hao Zhang, Kun Zhang, Shuigeng Zhou, Jihong Guan, Ji Zhang6538-6546

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that our method performs better in testing independence, which makes Re CIT (Zhang, Zhou, and Guan 2018) much faster and get a better performance in causal discovery. Performance Evaluation We first compare FRCIT with Re CIT (Zhang, Zhou, and Guan 2018) by extensive simulated experiments.
Researcher Affiliation Academia 1School of Computer, Guangdong University of Petrochemical Technology, China 2Shanghai Key Lab of Intelligent Information Processing, and School of Computer Science, Fudan University, China 3Department of Philosophy, Carnegie Mellon University, USA 4Department of Computer Science & Technology, Tongji University, China 5Zhejiang Lab, Hangzhou, China
Pseudocode Yes Algorithm 1 Fast regression based conditional independence test (FRCIT)
Open Source Code No No explicit statement or link to open-source code for the proposed FRCIT method is provided in the paper.
Open Datasets Yes We evaluate our method on real-world gene expression data in term of causal genes identification (Ruichu et al. 2013). This data (Golub and R. 1999) is a collection of 72 samples from leukemia patients...
Dataset Splits No For each parameter setting, we randomly repeat the testing 1000 times and average their results. (Explanation: The paper describes sample sizes and repetition for simulations, but does not provide specific training/validation/test splits or cross-validation details for the datasets used.)
Hardware Specification No No specific hardware details (such as GPU/CPU models or memory) used for running experiments are mentioned in the paper.
Software Dependencies No No specific software dependencies with version numbers (e.g., library names like PyTorch 1.9, or specific solver versions) are mentioned.
Experiment Setup Yes The significance level of Re CIT is fixed at α = 0.05. We check how the errors change when increasing the number of si (with aibi , 0) and the sample size n. For each parameter setting, we randomly repeat the testing 1000 times and average their results. ... We generate x and y according to the linear non-Gaussian SEM data generating procedure: x = Pl i=1 ai si and y = Pl i=1 bi si where ai, bi U( 1, 0.2) U(0.2, 1) are different for x and y, si is i.i.d. sampled from U( 0.5, 0.5)