tsai

Why we included this project

Anyone building predictive models on time-stamped data, whether from sensors, energy grids, finance, or hospital monitors, will find a lot to work with here. tsai collects the main modern deep learning architectures for time series, from RNN variants and CNNs to attention models like PatchTST, and handles classification, regression, forecasting, and imputation through a single API built on PyTorch and fastai. Instead of reimplementing each architecture yourself, you can swap models and run them against the same data pipeline, and the bundled datasets and tutorials lower the barrier for people new to the field. Walk-forward cross-validation and sklearn-style transforms give you honest evaluations before a model reaches production. It is a practical foundation for teams that want to move quickly on time series problems without writing all the boilerplate themselves.