Deep learning library for time series classification, regression, and forecasting in PyTorch.
Build, train, and deploy advanced neural networks for sequential data using PyTorch and fastai.
tsai is an open-source deep learning package designed specifically for time series and sequence tasks like forecasting, classification, and regression. It suits data scientists and machine learning engineers working on predictive supply chain analytics who need modern architectures like PatchTST or InceptionTime.
The library integrates with PyTorch and fastai to provide dozens of state-of-the-art architectures, built-in datasets, and sklearn-type pipeline transforms. It supports complex workflows including multi-class multivariate classification, walk-forward cross validation, and long-term forecasting.
It suits experienced machine learning practitioners with Python skills who need to build custom deep learning models for sequential operational data. It does not suit teams looking for a no-code forecasting dashboard or traditional statistical tools without deep learning requirements.
Access cutting-edge models like PatchTST, InceptionTime, MiniRocket, and various Transformers out of the box.
Quickly download and experiment with dozens of univariate, multivariate, regression, and forecasting datasets.
Utilize sklearn-type pipeline transforms and walk-forward cross validation specifically designed for time series.
Built on top of PyTorch and fastai to leverage modern hardware acceleration and training loops.
Free under its Apache-2.0 licence, with no paid hosted option provided by the maintainers. The only cost is running your own compute infrastructure for training deep learning models.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
tsai requires Python 3.10 or newer.
Yes, tsai supports time series forecasting, classification, regression, and imputation tasks.
See Zekai first in Google