gluonts

This probabilistic approach is critical for business decision-making where understanding forecast uncertainty confidence intervals, tail risks, and scenario distributions matters as much as the point forecast itself.

The library includes implementations of DeepAR (Amazon's production forecasting model), Temporal Fusion Transformer, NBEATS, WaveNet-based models, and Gaussian process approaches, all integrated into a consistent training and evaluation framework with standardized dataset loading, backtesting utilities, and probabilistic metric computation (CRPS, quantile loss, energy score).

GluonTS supports both MXNet and PyTorch backends through a framework-agnostic model API, and integrates with Hugging Face for Chronos time series foundation model inference.

GluonTS is open-source under the Apache 2.0 license and developed by the Amazon Research team responsible for AWS forecasting services.

It is used in production at Amazon for demand forecasting, capacity planning, and financial forecasting, and is widely adopted in academic time series research and industry forecasting applications in retail, energy, finance, and operations.

The library includes standard benchmark datasets and reproducible baselines that make it straightforward to compare new forecasting methods against established models.