gluonts: a probabilistic time series modeling library based on PyTorch

What it solves

GluonTS provides a framework for probabilistic time series modeling. It addresses the challenge of making predictions that aren't just single-point estimates, but rather probability distributions, allowing users to understand the uncertainty and prediction intervals of their forecasts.

How it works

Built on PyTorch, the library focuses on deep learning-based models (such as DeepAR) to analyze time series data. It allows users to load data via Pandas, split it into training and testing sets, and train estimators to generate probabilistic forecasts that can be visualized as shaded prediction intervals.

Who it’s for

Data scientists and researchers who need to perform neural time series forecasting and probabilistic modeling in Python.

Highlights

• Probabilistic Forecasting: Generates predictions as probability distributions rather than single values.
• Deep Learning Integration: Specifically designed for deep learning models based on PyTorch.
• Pandas Integration: Simplifies data loading and preparation using PandasDataset.
• Academic Foundation: Backed by multiple scientific publications in JMLR and arXiv.