PaddleTS 0.2.0 Release Note EN

API

• AutoTS for automatic hyper-parameter tuning is now introduced into PaddleTS.
• ReprBaseModel for building time series representation model is now introduced into PaddleTS.
• make_ts_transform & make_ml_model API supports third-party (e.g scikit-learn) data transformations and model integration.
• TSDataset supports representing & printing predicted output for probability forecasting models.
• Backtest new features: progress printing & probabilistic model adaptation & support returning both the score and the predictions at the same time.
• A cross-validation tool has been added to Utils, which can be combined with the new splitter tool in TSDataset to implement cross-validation.

Models

Time series probability forecasting and representation learning are now availalbe in PaddleTS. Meanwhile, PaddleTS has introduced 3 state-of-the-art deep learning models, Informer for time series forecasting, DeepAR for probability forecasting and TS2Vec for representation learning.

• Informer
• DeepAR
• TS2Vec

GPU Support

PaddleTS published a tutorial to guide developers run on GPU devices. Meanwhile, PaddleTS also have pre-built gpu-capable docker image:

registry.baidubce.com/paddlets/paddlets:0.2.0-gpu-paddle2.3.2-cuda11.2-cudnn8

Environment Setup

Pre-built cpu-capable docker image:

registry.baidubce.com/paddlets/paddlets:0.2.0