Boosted Tree Models

Use these guides for row-level boosted tree models. If the target is a regular time series, switch to the forecasting guides. If the learned ID representation itself is the point of the experiment, switch to the neural guides.

When To Use Boosting

• You need a tree model for numeric, class, or grouped-ranking targets.
• The signal may depend on time, location, membership, or residual structure.
• You want one entry point for fit, predict, save, load, and parameter selection, then a narrower page for each estimator.

Guides

• CartoBoost Regressor: numeric targets such as duration, fare, demand, or residuals.
• CartoBoost Classifier: binary or multiclass labels.
• CartoBoost Ranker: grouped candidate ordering.

Prepare And Validate The Data

Start from dense measured features, then add specialized controls only when they match the question:

• Python Estimator for fit, predict, save, and load behavior.
• Parameters for split policies, losses, fuzzy routing, and leaves.
• Categorical Features for native categorical and ordinal columns.
• Spatial CV Best Practices for blocked validation.
• Spatial Modeling for coordinates and spatial split behavior.
• Feature Schema for validation contracts.
• Sparse Features for zones, cells, and route memberships.

Compare against serious baselines on the same split, keep the feature access equal, and record the exact command, dataset, split, and metric summary.