AutoGluon-TimeSeries: Every Time Series Forecasting Model In One Library | by Nikos Kafritsas | Jan, 2024

The open-source panorama for time sequence is heating up.

This contains profitable libraries reminiscent of Darts, GluonTS, and Nixtla.

Final 12 months, Amazon constructed an extension of its AutoGluon library, which focuses on time-series — often called AutoGluon-Timeseries(AG-TS)[1].

AG-TS leverages the experience of different libraries:

• From Amazon itself (AutoGluon and GluonTS).
• From Nixtla (StatsForecast and MLForecast).

And the very best half: AG-TS has a user-friendly API— we will get predictions with just a few strains of code!

This text explores AG-TS and descriptions its capabilities. We will even assemble a easy mission, using the extensively identified Tourism dataset[2].

Let’s dive in

I’ve launched AI Horizon Forecast, a e-newsletter specializing in time-series and revolutionary AI analysis. Subscribe here to broaden your horizons!

AutoGluon–TimeSeries is an AutoML time-series framework, specializing in probabilistic forecasting and leveraging ensembling.

AG–TS helps:

• Open-Supply: The code is out there on GitHub — the library is a part of Amazon’s Autogluon suite.
• Consumer-Pleasant API: Load your information, and name the match() and predict() strategies.
• Mannequin selection: Entry to SOTA forecasting fashions throughout numerous classes, together with statistical fashions, tree-based approaches, and deep-learning fashions.
• Highly effective Ensembles: AG–TS incorporates automated configurations for optimum ensembling, a key technique in forecasting.
• Probabilistic output: The customers can generate level forecasts, with optionally available prediction intervals.
• Superior Efficiency: AG–TS excels in an analysis of 29 benchmark datasets, outperforming many forecasting strategies.