AutoGluon
Most AutoML systems treat the problem as a search — explore the space of models and settings, and hope the best one surfaces before time runs out. AutoGluon takes the opposite bet: train a set of models already known to be good, and combine them well.
How it works
AutoGluon is an open-source library from AWS. For tabular data it fits a fixed portfolio of strong model types in a sensible order — gradient-boosted trees (LightGBM, CatBoost, XGBoost), random forests, extremely randomised trees, nearest neighbours and neural networks — rather than tuning any single one exhaustively. It then bags each with repeated k-fold splits and stacks them in layers, passing out-of-fold predictions upward into a final weighted ensemble. Hyperparameter optimisation is optional and off the critical path; the accuracy comes from the ensembling. Models that overrun the budget are simply dropped, so a run always returns something usable.
Using it
from autogluon.tabular import TabularPredictor
predictor = TabularPredictor(label='churn').fit(train_data, presets='best_quality')
predictor.leaderboard(test_data)Raw columns go in — no imputation, scaling or encoding on your side. Presets trade accuracy against time, and the leaderboard shows every model trained with its score and fit time, which makes the result inspectable rather than magical. Beyond tables, AutoGluon handles images, text and mixed inputs through its multimodal predictor, and forecasting through its time-series predictor.
AutoGluon is an open-source AutoML library from AWS that produces strong models from raw tabular, text, image or time-series data in a few lines of code. Rather than searching hyperparameters exhaustively, it trains a fixed portfolio of proven models and combines them through bagging and multi-layer stacking. It matters because this ensemble-first design delivers top-tier tabular accuracy within a fixed time budget and needs almost no preprocessing from the user.
Say you have a table of data and a question you'd like answered, and you'd rather not learn a whole field first. AutoGluon is a free tool that takes your file more or less as it is — messy columns, missing values, text, dates — and in a few lines of code hands back a working prediction model.
Its trick is not to hunt for one perfect method. It trains a set of methods already known to work well, then blends their answers together, which normally beats any single one of them. You decide how much time it gets; it gives you the best it can manage in that window, along with a scoreboard showing how each model did.