合成时间序列数据集
收藏资源简介:
合成时间序列数据集是由摩根大通AI研究团队开发,用于评估大型语言模型在时间序列理解能力上的表现。该数据集包含9个子集,每个子集有200个时间序列样本,总共有1800个样本。每个样本的时间序列长度随机选择在30到150之间,以模拟不同长度的真实时间序列数据。数据集涵盖了多种时间序列特征,如趋势、季节性、波动性等,旨在为评估模型对时间序列特征的检测和分类能力提供一个全面的基准。该数据集的应用领域广泛,包括但不限于金融、医疗、气候和能源等,旨在解决时间序列分析中的自动化和精确性问题。
This synthetic time series dataset was developed by the JPMorgan Chase AI Research team to evaluate the performance of large language models (LLMs) on time series understanding tasks. The dataset consists of 9 subsets, each containing 200 time series samples, resulting in a total of 1800 samples across all subsets. The length of the time series for each sample is randomly selected between 30 and 150, aiming to simulate real-world time series data with varying lengths. The dataset covers a variety of time series characteristics, including trends, seasonality, volatility, and others, aiming to provide a comprehensive benchmark for evaluating models' ability to detect and classify time series features. This dataset has a wide range of application scenarios, including but not limited to finance, healthcare, climate, energy and other fields, and is designed to address the issues of automation and accuracy in time series analysis.

- 1Evaluating Large Language Models on Time Series Feature Understanding: A Comprehensive Taxonomy and Benchmark摩根大通AI研究 · 2024年



