GluonTS

GluonTS is a Python library for data scientists and researchers who need forecasts that express uncertainty. Its models predict a probability distribution over future values, so you can examine prediction intervals alongside the forecast. You train and use models within your own Python environment.

The library focuses on deep learning, with PyTorch and MXNet backends. Its model selection includes DeepAR, PatchTST, Temporal Fusion Transformer and WaveNet, alongside simpler approaches such as seasonal naive forecasting. That gives you baseline methods and neural models to compare within the same forecasting toolkit.

GluonTS also covers the work around a model. It accepts pandas DataFrames, supports training and test splits, and includes backtesting and evaluation metrics for checking predictions against held-out observations. Synthetic data generation supports experiments, while multivariate and hierarchical forecasting tools address related series and grouped data.

For researchers developing their own methods, it supports custom PyTorch models, training callbacks and model tuning with Optuna. Integrations include Prophet, StatsForecast and R forecasting tools. Forecast objects can plot predicted values with shaded uncertainty intervals.

GluonTS is open source under the Apache 2.0 license. Its scope includes both model development and forecast evaluation, with dataset handling and time-based features available in the same package. The current PyTorch installation uses gluonts[torch] and documents Python 3.10–3.14. Older MXNet-backed models require a separately compatible dependency environment.