Time Series Forecasting¶
For time-series data containing multiple individual series, AutoGluon can produce forecasting models to predict future values of each series based on historical observations of both this series and all of the other series in the dataset. A single call to AutoGluon TimeSeriesPredictor’s fit() automatically trains multiple models on a dataset containing multiple time-series measured over the same time period, and does not require you to manually deal with cumbersome issues like data cleaning, hyperparameter optimization, model selection, etc.
Most neural network-based models are from the GluonTS library. Allowed to contain missing values and additional (non-time-varying) static features, the data can be loaded from: a CSV file or the GluonTS format. AutoGluon-TimeSeries also supports simpler time series models such as exponential smoothing or ARIMA, through the statsmodels library.
Quick start tutorial on fitting models with time series datasets.
Detailed discussion of the time series forecasting capabilities in AutoGluon.
List of available forecasting models in AutoGluon-TimeSeries.
Frequently asked questions about AutoGluon-TimeSeries.