CausalTime
收藏资源简介:
CausalTime数据集是由清华大学自动化系的研究团队开发的,旨在为时间序列因果发现算法提供一个真实的测试平台。该数据集包含三个子集:空气质量指数(AQI)、交通流量和医疗记录,每个子集都基于真实数据生成,并附带真实的因果图。AQI子集包含中国多个城市的PM2.5污染指数数据,交通子集来自旧金山湾区,医疗子集则来自MIMIC-4数据库。数据集的生成过程涉及深度神经网络和归一化流技术,确保数据的真实性和复杂性,适用于评估和改进时间序列因果发现算法。
The CausalTime Dataset was developed by the research team from the Department of Automation, Tsinghua University, aiming to provide a realistic testbed for time-series causal discovery algorithms. This dataset includes three subsets: Air Quality Index (AQI), traffic flow, and medical records. Each subset is generated based on real-world data and is accompanied by a ground-truth causal graph. The AQI subset contains PM2.5 pollution index data from multiple cities in China; the traffic subset is sourced from the San Francisco Bay Area; the medical subset is derived from the MIMIC-4 database. The dataset generation process involves deep neural networks and normalizing flow techniques, ensuring the authenticity and complexity of the data, making it suitable for evaluating and improving time-series causal discovery algorithms.




