Time Series Library (TSLib)
收藏资源简介:
Time Series Library (TSLib)是由清华大学软件学院开发的一个综合性时间序列分析数据集,涵盖了30个不同领域的数据集,支持预测、分类、插补、异常检测等多种分析任务。数据集的创建过程结合了传统统计方法和先进的深度学习技术,旨在捕捉时间序列数据中的复杂时变模式和非线性关系。该数据集主要应用于金融风险评估、能源可持续性和天气预报等实际场景,旨在通过提供一个公平的基准来评估和识别现有模型的有效范围。
Time Series Library (TSLib) is a comprehensive time series analysis dataset developed by the School of Software, Tsinghua University. It covers 30 datasets across distinct domains, and supports various analytical tasks including forecasting, classification, imputation, and anomaly detection. The construction of this library integrates traditional statistical methods and cutting-edge deep learning technologies, aiming to capture complex time-varying patterns and nonlinear relationships within time series data. This dataset is primarily applied in practical scenarios such as financial risk assessment, energy sustainability, and weather forecasting, with the goal of providing a fair benchmark for evaluating existing models and identifying their effective application scopes.
Time Series Library (TSLib) 数据集概述
数据集简介
TSLib 是一个开源库,专为深度学习研究人员设计,特别是用于深度时间序列分析。该库涵盖了五种主流任务:长期和短期预测、填补、异常检测和分类。
最新动态
- 2024.07:发布了一篇关于深度时间序列模型的全面调查论文,基于TSLib进行严格的基准测试。
- 2024.04:将著名的序列模型Mamba纳入库中。
- 2024.03:将长期预测排行榜分为Look-Back-96和Look-Back-Searching两类。
- 2023.10:添加了iTransformer的实现,这是长期预测的最新模型。
- 2023.09:为TimesNet和该库添加了详细的教程,适合深度时间序列分析的初学者。
- 2023.02:发布了TSLib,作为时间序列模型的综合基准和代码库。
排行榜
截至2024年3月,五个不同任务的顶级模型如下:
| 排名 | 长期预测<br>Look-Back-96 | 长期预测<br>Look-Back-Searching | 短期预测 | 填补 | 分类 | 异常检测 |
|---|---|---|---|---|---|---|
| 🥇 1st | iTransformer | TimeMixer | TimesNet | TimesNet | TimesNet | TimesNet |
| 🥈 2nd | TimeMixer | PatchTST | Non-stationary Transformer | Non-stationary Transformer | Non-stationary Transformer | FEDformer |
| 🥉 3rd | TimesNet | DLinear | FEDformer | Autoformer | Informer | Autoformer |
已包含的模型
以下模型已包含在TSLib中,并提供了代码链接:
- TimeMixer
- TSMixer
- iTransformer
- PatchTST
- TimesNet
- DLinear
- LightTS
- ETSformer
- Non-stationary Transformer
- FEDformer
- Pyraformer
- Autoformer
- Informer
- Reformer
- Transformer
- Mamba
- SegRNN
- Koopa
- FreTS
- TiDE
- FiLM
- MICN
- Crossformer
- TFT
使用方法
-
安装Python 3.8并执行以下命令安装依赖:
pip install -r requirements.txt
-
准备数据,可以从Google Drive或Baidu Drive获取预处理好的数据集,并放置在
./dataset文件夹中。 -
训练和评估模型,提供了实验脚本在
./scripts/文件夹下,例如:bash ./scripts/long_term_forecast/ETT_script/TimesNet_ETTh1.sh
-
开发自己的模型,将模型文件添加到
./models文件夹,并在Exp_Basic.model_dict中包含新模型,创建相应的脚本。
引用
如果发现该库有用,请引用相关论文:
@inproceedings{wu2023timesnet, title={TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis}, author={Haixu Wu and Tengge Hu and Yong Liu and Hang Zhou and Jianmin Wang and Mingsheng Long}, booktitle={International Conference on Learning Representations}, year={2023}, }
@article{wang2024tssurvey, title={Deep Time Series Models: A Comprehensive Survey and Benchmark}, author={Yuxuan Wang and Haixu Wu and Jiaxiang Dong and Yong Liu and Mingsheng Long and Jianmin Wang}, booktitle={arXiv preprint arXiv:2407.13278}, year={2024}, }
联系
如有任何问题或建议,请联系维护团队或通过Issues描述。

- 1Deep Time Series Models: A Comprehensive Survey and Benchmark清华大学软件学院 · 2024年



