Auto Gluon

Auto Gluon automates the entire machine learning pipeline, including model training, hyperparameter tuning, and model selection. It offers an efficient solution for forecasting the future values using historical data and relevant covariates. It automatically trains multiple models to generate accurate probabilistic forecasts. User don’t have to worry about complex tasks like model selection or hyperparameter tuning.

AutoGluon leverages a combination of state-of-the-art forecasting algorithms. It integrates traditional statistical methods like ETS and ARIMA from StatsForecast, powerful tree-based models such as LightGBM from AutoGluon-Tabular, and advanced deep learning models like DeepAR and Temporal Fusion Transformer from GluonTS. Additionally, it includes Chronos, a pretrained zero-shot forecasting model, providing robust and adaptable predictions across various time series data.

Key Features

• Multiple Model Support: Combines traditional statistical models (ARIMA, ETS) with machine learning models (LightGBM) and deep learning models (DeepAR, Transformer-based models).

• Probabilistic Forecasting: Generates uncertainty estimates for forecasts using probabilistic models.

• Zero-Shot Forecasting: Supports pretrained models like Chronos for instant forecasting without additional training.

• Automatic Model Selection: Selects the best model using automated hyperparameter tuning and ensembling.

AutoGluon Mandatory Parameters

Param Name	Param Description	Default Values	Possible Values
EVALMETRIC	Defines the evaluation metric used in the model.	RMSE
HYPERPARAMETERS	Lists the hyperparameters used for model tuning.	["SeasonalNaive", "NPTS", "Zero", "AutoETS", "AutoARIMA", "RecursiveTabular", "DirectTabular"]