gluon-ts

What it is

A Python package for probabilistic time series modeling, focusing on deep learning based models built on PyTorch. It trains models such as DeepAR on historical series and produces forecasts expressed as probability distributions with prediction intervals, rather than single-point predictions. Quants working on forecasting components of trading strategies may find it useful, and the project is actively maintained rather than archived. Its emphasis on deep learning models means users should expect a research-oriented toolkit rather than a turnkey product, and effective use typically requires familiarity with probabilistic forecasting concepts.