Directed Acyclic Graph Structure Learning from Dynamic Graphs

Authors: Shaohua Fan, Shuyang Zhang, Xiao Wang, Chuan Shi

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive simulation experiments with broad range settings which may encounter in real world, validating the effectiveness of our approach in revealing the feature generation mechanism of dynamic graphs. The experiments on real-world datasets well demonstrate the rationality of the relationships inferenced by Graph NOTEARS.
Researcher Affiliation Academia 1Beijing University of Posts and Telecommunications, China 2Peng Cheng Laboratory, China {fanshaohua, sonyazhang, xiaowang, shichuan}@bupt.edu.cn
Pseudocode Yes For the detailed pseudocode of Eq. (1), please refer to Appendix A.2
Open Source Code Yes Code and data: https://github.com/googlebaba/Graph NOTEARS.
Open Datasets Yes Code and data: https://github.com/googlebaba/Graph NOTEARS.
Dataset Splits No The paper describes a temporal splitting strategy ("we use the first T-1 timestamps to predict last T-p timestamps") which serves as a form of train/test split for time-series data, but it does not explicitly provide details for a separate validation split, nor specific percentages or sample counts for general training, validation, and test datasets in the conventional sense.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory, cloud instances) used for running the experiments.
Software Dependencies No The paper does not specify particular software dependencies or library versions used in the experiments (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes For all methods, we set hyperparameters λW = λP = 0.01. For the weight thresholds, following (Zheng et al. 2018), we choose τW = τP = 0.3 for all the methods.