AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data

By Nick Erickson, Jonas W. Mueller, Alexander Shirkov, Hang Zhang, Pedro Larroy, Mu Li, Alex Smola. 2020

995 citations recorded by the source corpus.

We introduce AutoGluon-Tabular, an open-source AutoML framework that requires only a single line of Python to train highly accurate machine learning models on an unprocessed tabular dataset such as a CSV file. Unlike existing AutoML frameworks that primarily focus on model/hyperparameter selection, AutoGluon-Tabular succeeds by ensembling multiple models and stacking them in multiple layers. Experiments reveal that our multi-layer combination of many models offers better use of allocated training time than seeking out the best. A second contribution is an extensive evaluation of public and commercial AutoML platforms including TPOT, H2O, AutoWEKA, auto-sklearn, AutoGluon, and Google AutoML Tables. Tests on a suite of 50 classification and regression tasks from Kaggle and the OpenML AutoML Benchmark reveal that AutoGluon is faster, more robust, and much more accurate. We find that AutoGluon often even outperforms the best-in-hindsight combination of all of its competitors. In two popular Kaggle competitions, AutoGluon beat 99% of the participating data scientists after merely 4h of training on the raw data.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

VEGA: Towards an End-to-End Configurable AutoML Pipeline (2020, methodology): “AutoGluon (Erickson et al., 2020) is another AutoML toolkit and can be used for tabular prediction, image classiﬁcation, object detection, and text prediction.”
Best Practices for Scientific Research on Neural Architecture Search (2019, methodology): “Although several well-engineered and flexible AutoDL packages were recently proposed (Jin et al., 2019; O’Malley et al., 2019; Erickson et al., 2020; Zimmer et al., 2020), their main goal is ease of use, rather than allowing clean comparisons of several NAS methods.”
TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second (2022, methodology): “As more complex but also potentially powerful baselines, we chose three AutoML systems: AutoGluon [8], which combines ML models including neural networks and tree-based models into a stacked ensemble, Auto-sklearn 2.”
Auto-Sklearn 2.0: Hands-free AutoML via Meta-Learning (2020, background): “Lastly, AutoGluon takes a radically different approach and completely drops hyperparameter optimization and invests all available time into building a robust stacking model (Erickson et al., 2020).”
Amazon SageMaker Autopilot: a white box AutoML solution at scale (2020, methodology): “AutoGluon-Tabular (Erickson et al., 2020) uses multilayer stacking with K-fold bagging over gradient boosting, residual MLP, linear and KNN base learners, whose hyperparameters are set to defaults.”
Accurate predictions on small data with a tabular foundation model (2025, positive): “For TabPFN (PHE) and AutoGluon, we start with a minimal budget of 300 s for tuning because AutoGluon otherwise does not reliably return results.”
Hyperparameter optimization: Foundations, algorithms, best practices, and open challenges (2021, positive): “H2O automl (LeDell & Poirier, 2020) and AutoGluon (Erickson et al., 2020) train a set of base models and ensemble them with a single layer (H2O) or multi-layer stacks (AutoGluon).”
SAINT: Improved Neural Networks for Tabular Data via Row Attention and Contrastive Pre-Training (2021, methodology): “D Baselines We use Auto-Gluon2 [Erickson et al., 2020] framework to run the train the traditional models as well as to perform the hyperparameter sweep.”
OpenML Benchmarking Suites (2017, background): “Erickson et al. [2020] use additional datasets from Kaggle competitions to compare directly to solutions proposed by human competitors.”
Fast, Accurate, and Simple Models for Tabular Data via Augmented Distillation (2020, methodology): “We apply FAST-DAD to large stack-ensemble predictors from AutoGluon [1] to produce individual models that are over 10,000ˆ faster than AutoGluon and over 10ˆ faster, yet still more accurate, than ensemble predictors produced by H2O-AutoML [7] and AutoSklearn [2].”
FLAML: A Fast and Lightweight AutoML Library (2019, background): “In addition, ensemble of multiple learners is often considered useful for boosting accuracy at the cost of increased inference latency (Erickson et al., 2020).”