AI GitHub
LightGBM for efficient tree boosting
Train gradient-boosted trees using histogram-based splitting and configurable leaf growth.
Key Takeaways
- LightGBM groups feature values for split decisions and grows trees by selecting leaves to split.
- Define categorical and numerical features explicitly in the training dataset.
- Train a small baseline with explicit categorical handling.
- Leaf-wise growth can fit small samples too closely.
How it works
LightGBM groups feature values for split decisions and grows trees by selecting leaves to split. Those choices make feature representation and leaf constraints important on real datasets.
What you need
Define categorical and numerical features explicitly in the training dataset. Keep validation splits fixed when comparing training speed, memory and error.
What to try first
Train a small baseline with explicit categorical handling. Compare performance, memory and error by subgroup against the current model.
Keep training and serving feature definitions together. A categorical value converted differently at inference can change predictions without raising an error.
Where to look for failures
Leaf-wise growth can fit small samples too closely. Validate leaf constraints and preserve feature order and category encoding when serving.
Related Franklin pages
Keep data splits, features and metric settings fixed when comparing tabular baselines. Measure model size and serving time alongside predictive performance.
- Train and compare candidate models using documented presets and resource budgets. AutoGluon for tabular and multimodal baselines
- Build predictive pipelines with consistent preprocessing, validation and estimator APIs. scikit-learn for reproducible machine-learning baselines
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 ↗