Releases: GestaltCogTeam/BasicTS
Releases · GestaltCogTeam/BasicTS
v1.1.0
🚀 New Features
- Models: Upgraded FiLM, TiDE, and Koopa
- Optimized the user experience of configuration files
🐛 Bug Fixes
- Fixed the issue where the launcher reported errors when using DDP for training
- Fixed issues with the BLAST dataset
- Fixed numerous problems in models
- Fixed the issue where configurations could not be loaded from JSON files when using selective learning
- Fixed the issue where assignments in configurations could be unexpectedly overwritten by default values
🥂 New Contributors
🚀 新功能
- 模型：升级了FiLM、TiDE、Koopa
- 优化了配置文件的使用体验
🐛 修复问题
- 修复了launcher使用DDP训练报错的问题
- 修复了BLAST数据集的问题
- 修复了大量模型中存在的问题
- 修复了使用选择学习无法从JSON文件中加载配置的问题
- 修复了配置中的赋值会意外被默认值覆盖的问题
🥂 新贡献者
Contributors
wgawmy
Assets 4
Version 1.0.2
🚀 New Features
- Configuration: Optimized design. Users can now pass arbitrary parameters when constructing configuration classes.
- Models: Added FITS (ICLR'24), FreTS (NeurIPS'23), Leddam (ICML'24), and upgraded StemGNN.
- Modules: Added the normalization component STNorm (SIGKDD'21).
🐛 Bug Fixes
- Fixed an issue where the runner could not properly resume training.
- Fixed parameter inconsistencies between configuration classes and actual inputs.
- Resolved issues in Crossformer and LightTS.
- Fixed an issue where the LRScheduler could not run properly.
🚀 新功能
- 配置：优化设计，用户现在可以在构造配置类时传递任意参数
- 模型：新增FITS (ICLR'24)、FreTS (NeurIPS'23)、Leddam (ICML'24)，升级了StemGNN
- 模块：新增归一化组件 STNorm (SIGKDD'21)
🐛 修复问题
- 修复了runner无法正常续训的问题
- 修复配置类中参数与实际传入不一致的问题
- 修复Crossformer、LightTS中存在的问题
- 修复了使用LRScheduler时无法正常运行的问题
Version 1.0.1
🚀 New Features
- Support loading configurations from JSON files
- Provide runnable demos for various tasks in the
examplesdirectory - Upgraded SegRNN and Non-Stationary Transformer (upgrades for other models are in progress)
- Optimized the architecture of Transformer components in modules for better extensibility
- Added ProbAttention (AAAI'21 Informer) to the attention components in modules
🐛 Bug Fixes
- Fixed bugs in Autoformer, Crossformer, and Informer models
- Fixed issues in the classification and imputation task pipelines
🚀 新功能
- 支持从JSON文件加载配置
- 在项目
examples路径下提供了各任务可运行的demo - 升级了SegRNN、Non-Stationary Transformer (其他模型的升级正在路上)
- 优化了模块中Transformer组件的架构，使之更具可扩展性
- 在模块的注意力组件中加入了ProbAttention (AAAI'21 Informer)
🐛 修复问题
- 修复了Autoformer、Crossformer、Informer模型中存在的bug
- 修复了分类与插补任务流程中存在的bug
Version 1.0.0
📣 Important
BasicTS version 1.0 is here! BasicTS 1.0 has undergone a comprehensive upgrade focusing on user-friendliness, multi-task support, and extensibility.
🚀 New Features
- Optimized user experience, enabling model training or evaluation with just three lines of Python code
- Added directly callable time series deep learning modules (Transformer components, MLP, ...)
- Support for time series classification and imputation tasks
- Enhanced extensibility with new taskflow and callback mechanisms, allowing users to modify workflows without changing the runner
🐛 Bug Fixes
- Fixed the issue of potentially high GPU memory usage during model evaluation
- Fixed incorrect calculation of training and evaluation metrics
🔧 Upgrade Instructions
For BasicTS 1.0 version, there's no need to clone the repository anymore. Simply download basicts-1.0-py3-none-any.whl and execute:
pip install basicts-1.0-py3-none-any.whl📣 重要消息
BasicTS 1.0 版本来了！BasicTS 1.0 针对用户友好，多任务支持和可扩展性方面进行了全面升级。
🚀 新功能
- 优化了用户的使用体验，三句Python代码实现模型训练或评估
- 增加用户可以直接调用的时序深度学习模块（Transformer components, MLP, ...）
- 支持时间序列分类和插补任务
- 优化了可扩展性，新增taskflow和callback机制，用户无需再修改runner
🐛 修复问题
- 修复在GPU上评估模型显存可能占用过高的问题
- 修复训练、评估指标计算可能不正确的问题
🔧 升级说明
安装BasicTS v1.0 无需再clone仓库，只需下载basicts-1.0-py3-none-any.whl，然后执行：
pip install basicts-1.0-py3-none-any.whlVersion 1.0.0-beta
📣 Important
BasicTS version 1.0 is here! Welcome to experience the beta version first. BasicTS 1.0 has undergone a comprehensive upgrade focusing on user-friendliness, multi-task support, and extensibility.
🚀 New Features
- Optimized user experience, enabling model training or evaluation with just three lines of Python code
- Added directly callable time series deep learning modules (Transformer components, MLP, ...)
- Support for time series classification and imputation tasks
- Enhanced extensibility with new taskflow and callback mechanisms, allowing users to modify workflows without changing the runner
🐛 Bug Fixes
- Fixed the issue of potentially high GPU memory usage during model evaluation
- Fixed incorrect calculation of training and evaluation metrics
🔧 Upgrade Instructions
For BasicTS 1.0 beta version, there's no need to clone the repository anymore. Simply download basicts-1.0-py3-none-any.whl and execute:
pip install basicts-1.0-py3-none-any.whl📣 重要消息
BasicTS 1.0 版本来了！欢迎抢先体验测试版。BasicTS 1.0 针对用户友好，多任务支持和可扩展性方面进行了全面升级。
🚀 新功能
- 优化了用户的使用体验，三句Python代码实现模型训练或评估
- 增加用户可以直接调用的时序深度学习模块（Transformer components, MLP, ...）
- 支持时间序列分类和插补任务
- 优化了可扩展性，新增taskflow和callback机制，用户无需再修改runner
🐛 修复问题
- 修复在GPU上评估模型显存可能占用过高的问题
- 修复训练、评估指标计算可能不正确的问题
🔧 升级说明
安装BasicTS v1.0 beta无需再clone仓库，只需下载basicts-1.0-py3-none-any.whl，然后执行：
pip install basicts-1.0-py3-none-any.whl