Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GluonTS: Probabilistic Time Series Models in Python

Alexander Alexandrov, Konstantinos Benidis, Michael Bohlke-Schneider, Valentin Flunkert, Jan Gasthaus, Tim Januschowski, Danielle C. Maddix, Syama Rangapuram, David Salinas, Jasper Schulz, Lorenzo Stella, Ali Caner Türkmen, Yuyang Wang

TL;DR

GluonTS provides a modular, MXNet Gluon-based toolkit for probabilistic time series modeling, integrating data I/O, flexible distributions, forecast objects, evaluation, and model-building primitives to enable rapid, reproducible prototyping of forecasting and anomaly-detection methods. It covers generative, discriminative, and auto-regressive approaches, including DeepState, DeepAR, Transformer, and quantile-decoding variants, with support for both real and synthetic data streams. The paper demonstrates a comprehensive benchmarking suite on 11 public datasets and emphasizes reproducibility through logging and JSONable configurations. This work offers a practical platform for researchers and practitioners to develop, benchmark, and deploy advanced time series models in a probabilistic framework.

Abstract

We introduce Gluon Time Series (GluonTS, available at https://gluon-ts.mxnet.io), a library for deep-learning-based time series modeling. GluonTS simplifies the development of and experimentation with time series models for common tasks such as forecasting or anomaly detection. It provides all necessary components and tools that scientists need for quickly building new models, for efficiently running and analyzing experiments and for evaluating model accuracy.

GluonTS: Probabilistic Time Series Models in Python

TL;DR

GluonTS provides a modular, MXNet Gluon-based toolkit for probabilistic time series modeling, integrating data I/O, flexible distributions, forecast objects, evaluation, and model-building primitives to enable rapid, reproducible prototyping of forecasting and anomaly-detection methods. It covers generative, discriminative, and auto-regressive approaches, including DeepState, DeepAR, Transformer, and quantile-decoding variants, with support for both real and synthetic data streams. The paper demonstrates a comprehensive benchmarking suite on 11 public datasets and emphasizes reproducibility through logging and JSONable configurations. This work offers a practical platform for researchers and practitioners to develop, benchmark, and deploy advanced time series models in a probabilistic framework.

Abstract

We introduce Gluon Time Series (GluonTS, available at https://gluon-ts.mxnet.io), a library for deep-learning-based time series modeling. GluonTS simplifies the development of and experimentation with time series models for common tasks such as forecasting or anomaly detection. It provides all necessary components and tools that scientists need for quickly building new models, for efficiently running and analyzing experiments and for evaluating model accuracy.

Paper Structure

This paper contains 19 sections, 10 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Library design and components
  3. Data I/O and processing
  4. Predictor
  5. Distribution / Output
  6. Forecast object
  7. Evaluation
  8. Time Series Model Building
  9. Time Series Problems
  10. Time Series Models
  11. Generative Models
  12. Discriminative Models
  13. Auto-regressive models
  14. Experiments
  15. Related Work
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Histogram of forecast error measured in on traffic-dataset for ETS, Prophet and DeepAR.
  • Figure 2: Examples of anomalies detected using a trained DeepAR forecast model on the electricity dataset.
  • Figure 3: Running time in second for all datasets and methods (logarithmic scale). DeepAR is labeled as AR2N2 in this figure.