AutoTS

AutoTS is an open-source Python package for people who need forecasts across many time series without choosing a model and preprocessing method for each one. It runs in your own Python environment and uses automated search to compare forecasting approaches for your data. The license is MIT.

Its model range includes simple baseline forecasts, statistical methods, machine learning and deep learning. All models can forecast multiple series and return upper and lower bounds alongside predictions. Many also accept external predictors, such as other variables you expect to affect the forecast.

AutoTS searches for combinations of models, data transformations and ensembles using genetic algorithms. Its horizontal and mosaic ensembles can assign different models to different series, rather than require one model to handle the whole dataset. Most models support large collections of series, and parallel execution can use multiple CPU cores.

For teams already working with pandas, the models and transformations operate directly on DataFrames. AutoTS accepts both long and wide time series data, so you don't need to convert it into a package-specific object format. Its lower-level models and time series transformations also work independently of the automated search, with interfaces similar to scikit-learn.

The package includes an MCP server. Some forecasting models and methods require additional Python dependencies.