SELF: Structural Equational Likelihood Framework for Causal Discovery

Authors: Ruichu Cai, Jie Qiao, Zhenjie Zhang, Zhifeng Hao

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical evaluations using XGBoost validate the superiority of our proposal over state-of-the-art solutions, on both synthetic and real world causal structures. Experiment Settings To investigate the effectiveness and of genericity of SELF, the algorithms are tested on both linear non-Gaussian and nonlinear data, generated from synthetic and real world causal structures.
Researcher Affiliation Collaboration Ruichu Cai,1 Jie Qiao,1 Zhenjie Zhang,2 Zhifeng Hao1, 3 1 School of Computer Science, Guangdong University of Technology, China 2 Advanced Digital Sciences Center, Illinois at Singapore Pte. Ltd., Singapore 3 School of Mathmatics and Big Data, Foshan University, China
Pseudocode Yes Algorithm 1 Hill-Climbing Based Causal Structure Search
Open Source Code Yes The implementation of SELF can be found in CRAN 1. https://cran.r-project.org/web/packages/SELF/index.html
Open Datasets Yes The algorithms are tested on both linear non-Gaussian and nonlinear data, generated from synthetic and real world causal structures. we explore the performance of the algorithms on four frequently used real world structures (Scutari 2009). The statistics of the structures are given in Table 2.
Dataset Splits No The paper does not provide specific dataset split information for training, validation, or testing, such as percentages or sample counts. It describes ranges for number of samples for synthetic data generation, but not how they are split for model training and evaluation.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments. It only mentions using 'XGBoost' which is a software framework.
Software Dependencies No The paper mentions software like 'XGBoost-GBLinear', 'XGBoost-GBTree', 'bnlearn package in R', and 'Compare Causal Networks package'. However, it does not provide specific version numbers for these software components, which is required for reproducible description.
Experiment Setup No While the paper has an 'Experiment Settings' section that describes parameters for synthetic data generation (number of variables, samples, average indegree), it does not provide specific hyperparameters or system-level training settings for the models used (e.g., learning rate, batch size, number of epochs for XGBoost).