AI GitHub
AutoGluon for tabular and multimodal baselines
Train and compare candidate models using documented presets and resource budgets.
Key Takeaways
- AutoGluon exposes task predictors and can train multiple models or ensembles.
- Set a training budget and hold out a validation period or group when appropriate.
- Give a predictor a fixed time budget and an application-appropriate validation split.
- An ensemble can score well but be expensive to serve.
How it works
AutoGluon exposes task predictors and can train multiple models or ensembles. Presets control how much work the system does and influence training and inference costs.
What you need
Set a training budget and hold out a validation period or group when appropriate. Check the task predictor and preset match the data type and deployment plan.
What to try first
Give a predictor a fixed time budget and an application-appropriate validation split. Inspect the leaderboard and measure deployment latency for the selected model.
Check whether the chosen ensemble can run within the actual serving budget. The fastest acceptable model can be easier to deploy than the top row of a leaderboard.
Where to look for failures
An ensemble can score well but be expensive to serve. Keep preprocessing and evaluation stable when comparing presets, and check model artifact size.
Related Franklin pages
Keep data splits, features and metric settings fixed when comparing tabular baselines. Measure model size and serving time alongside predictive performance.
- Build predictive pipelines with consistent preprocessing, validation and estimator APIs. scikit-learn for reproducible machine-learning baselines
- Train boosted tree models for tasks such as tabular classification, regression and ranking. XGBoost for gradient-boosted trees
For a primer on learning from data and evaluating predictions, see Franklin’s glossary. Machine learning
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗