FoundTS
收藏资源简介:
FoundTS是一个用于时间序列预测的基础模型综合评估和比较的基准数据集。该数据集由华东师范大学创建,涵盖了多个领域和不同特征的时间序列数据,旨在全面评估和比较不同基础模型在时间序列预测中的表现。数据集包括来自股票、健康、能源、电力、环境、交通、自然、银行、网络和经济等十个领域的数据,具有季节性、趋势、平稳性等多种特征。FoundTS支持零样本、少样本和全样本等多种评估策略,通过标准化数据分割、加载、归一化和少样本采样等流程,确保评估的公平性和全面性。该数据集主要应用于时间序列预测领域,旨在解决现有模型在新领域或未见数据上泛化能力不足的问题。
FoundTS is a benchmark dataset for comprehensive evaluation and comparison of foundation models in time series forecasting. Developed by East China Normal University, this dataset covers time series data from multiple domains with diverse characteristics, aiming to thoroughly assess and compare the performance of various foundation models in time series forecasting. It includes data from ten domains such as stocks, healthcare, energy, power, environment, transportation, natural sciences, banking, networks, and economics, and possesses multiple characteristics including seasonality, trend, and stationarity. FoundTS supports multiple evaluation strategies such as zero-shot, few-shot, and full-sample settings, and ensures the fairness and comprehensiveness of evaluations through standardized workflows including data splitting, loading, normalization, and few-shot sampling. This dataset is primarily applied in the time series forecasting domain, with the objective of addressing the insufficient generalization ability of existing models on new domains or unseen data.

- 1FoundTS: Comprehensive and Unified Benchmarking of Foundation Models for Time Series Forecasting华东师范大学 · 2024年



